From 5c48fed1abcaa7533bf93c129a4de5064c5a3d69 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 03:45:48 +0300
Subject: [PATCH 01/11] =?UTF-8?q?Editor:=20shortcut=20Ideas=20bulb=20?=
 =?UTF-8?q?=E2=80=94=20quiet=20nudge=20for=20unused=20shortcut=20categorie?=
 =?UTF-8?q?s?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

A 💡 next to the scene prompt and Composer global prompt lights up when whole
shortcut categories (camera moves, sub-actions, …) exist in the library but are
absent from the prompt. Click-only (never auto-opens): a category-grouped
popover lists every missing category with count, random insertable samples,
+N-more browse per category, and a shuffle. Chips insert at the caret through
the field's normal input path. Toggleable via Settings > Editor > Shortcut ideas.

Guards: no popover off a hidden/detached bulb (stranded top-left window),
dismiss listeners detach on every close path, popover closes when a re-render
removes the bulb.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 movie_editor/frontend/composer.js        |   1 +
 movie_editor/frontend/editor_settings.js |   6 +-
 movie_editor/frontend/index.html         |   1 +
 movie_editor/frontend/inspector.js       |   7 +-
 movie_editor/frontend/store.js           |   2 +-
 movie_editor/frontend/styles.css         |  19 ++
 movie_editor/frontend/suggestions.js     | 223 +++++++++++++++++++++++
 7 files changed, 256 insertions(+), 3 deletions(-)
 create mode 100644 movie_editor/frontend/suggestions.js

diff --git a/movie_editor/frontend/composer.js b/movie_editor/frontend/composer.js
index d6d52d85..189f8ef0 100644
--- a/movie_editor/frontend/composer.js
+++ b/movie_editor/frontend/composer.js
@@ -213,6 +213,7 @@
       "Edits apply automatically and stay in sync with the timeline's per-scene prompts. Shortcuts expand at generation time."));
     composeTextarea = ta;
     if (window.ShortcutAutocomplete) window.ShortcutAutocomplete.attach(ta);
+    if (st.project && window.ShortcutSuggest) titleRow.append(window.ShortcutSuggest.bulb(ta));
 
     // Variables ($name) — collapsible, same toggle pattern as Settings ▸ Built-in pipeline.
     wrap.append(composeVariables());
diff --git a/movie_editor/frontend/editor_settings.js b/movie_editor/frontend/editor_settings.js
index 3074aef7..b77102a8 100644
--- a/movie_editor/frontend/editor_settings.js
+++ b/movie_editor/frontend/editor_settings.js
@@ -158,6 +158,10 @@
       "Prompt autocomplete",
       "Suggest matching shortcuts while you type in the global prompt and scene prompts. Shows the trigger, its prompt, and category.",
       "autocomplete"));
+    content.append(toggleRow(
+      "Shortcut ideas",
+      "A 💡 next to prompt fields lights up when shortcut categories in your library (camera moves, sub-actions…) aren't used in the prompt. Click it for insertable ideas — it never pops up on its own.",
+      "suggestions"));
     content.append(toggleRow(
       "Use anchor",
       "Text before the first split trigger is a shared anchor prepended to every scene. Turn off to make that leading text Scene 1 instead.",
@@ -169,7 +173,7 @@
   window.SettingsWindow.register({
     id: "editor", group: "", title: "Editor",
     subtitle: "Per-browser preferences — they apply to this browser, not the project.",
-    keywords: "autocomplete anchor prompt shortcuts i2v bypass guide preferences",
+    keywords: "autocomplete anchor prompt shortcuts ideas suggestions i2v bypass guide preferences",
     iconBg: "linear-gradient(180deg,#6aa9ff,#3b6fd9)",
     icon: '<svg viewBox="0 0 16 16" width="13" height="13" fill="none" stroke="#fff" stroke-width="1.5" stroke-linecap="round"><path d="M2 4.5h12M2 8h12M2 11.5h12"/><circle cx="6" cy="4.5" r="1.7" fill="#fff" stroke="none"/><circle cx="11" cy="8" r="1.7" fill="#fff" stroke="none"/><circle cx="5" cy="11.5" r="1.7" fill="#fff" stroke="none"/></svg>',
     mount,
diff --git a/movie_editor/frontend/index.html b/movie_editor/frontend/index.html
index 47061c3b..5e6f27cc 100644
--- a/movie_editor/frontend/index.html
+++ b/movie_editor/frontend/index.html
@@ -130,6 +130,7 @@
   <script src="editor_settings.js"></script>
   <script src="temp_browser.js"></script>
   <script src="autocomplete.js"></script>
+  <script src="suggestions.js"></script>
   <script src="slot_picker.js"></script>
   <script src="welcome.js"></script>
   <script src="pipeline_setup.js"></script>
diff --git a/movie_editor/frontend/inspector.js b/movie_editor/frontend/inspector.js
index aab48a90..f0711a0f 100644
--- a/movie_editor/frontend/inspector.js
+++ b/movie_editor/frontend/inspector.js
@@ -288,7 +288,12 @@
 
     const ta = el("textarea"); ta.rows = 4; ta.value = root.text || ""; ta.placeholder = "Describe this scene…"; ta.dataset.k = "sc-text";
     ta.oninput = () => S.patchSceneQuiet(scene.id, { text: ta.value });
-    body.append(field("Prompt", ta));
+    // Ideas bulb floats OUTSIDE the label — a button inside a <label> becomes its
+    // implicit control (double-fired clicks, stolen focus target).
+    const promptWrap = el("div", "sugg-fieldwrap");
+    promptWrap.append(field("Prompt", ta));
+    if (window.ShortcutSuggest) promptWrap.append(window.ShortcutSuggest.bulb(ta));
+    body.append(promptWrap);
     if (window.ShortcutAutocomplete) window.ShortcutAutocomplete.attach(ta);
     renderSourceField(st, root, scene);
     if (PC()?.usesChainSampler(st)) {
diff --git a/movie_editor/frontend/store.js b/movie_editor/frontend/store.js
index 69901e6b..39d7ff77 100644
--- a/movie_editor/frontend/store.js
+++ b/movie_editor/frontend/store.js
@@ -39,7 +39,7 @@
   // prompt is parsed/edited, not what is generated from already-distributed scenes.
   const EDITOR_SETTINGS_KEY = "funpack_editor_settings";
   const EDITOR_SETTINGS_DEFAULTS = {
-    autocomplete: true, anchorEnabled: true,
+    autocomplete: true, suggestions: true, anchorEnabled: true,
     // "Anchor as guide" i2v bypass: declare a custom i2v node so it can be forced
     // to a state (value-only) on anchor_guide runs. anchorGuideBypass = {node, input, value}.
     anchorGuideHasI2v: false, anchorGuideBypass: null,
diff --git a/movie_editor/frontend/styles.css b/movie_editor/frontend/styles.css
index a8642aa8..6dcc3da5 100644
--- a/movie_editor/frontend/styles.css
+++ b/movie_editor/frontend/styles.css
@@ -1561,6 +1561,25 @@ body.tour-mode { overflow: hidden; }
 .ac-prompt { color: var(--muted); font-size: 11.5px; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; flex: 1; min-width: 0; }
 .ac-cat { color: var(--teal); font-family: var(--font-mono); font-size: 9.5px; letter-spacing: .5px; text-transform: uppercase; flex-shrink: 0; }
 
+/* Shortcut ideas bulb + popover (quiet nudge — lights up when categories go unused) */
+.sugg-bulb { padding: 0 4px; font-size: 12px; line-height: 1; opacity: .35; filter: grayscale(1); transition: opacity .3s, filter .3s; }
+.sugg-bulb.on { opacity: 1; filter: none; text-shadow: 0 0 7px rgba(255, 196, 87, .55); }
+.sugg-fieldwrap { position: relative; }
+.sugg-fieldwrap .sugg-bulb { position: absolute; top: -3px; right: 0; }
+.sugg-pop { position: fixed; z-index: 99999; width: 330px; max-height: 280px; overflow-y: auto;
+  background: #16140f; border: 1px solid var(--line); border-radius: 8px; box-shadow: var(--shadow); padding: 6px; }
+.sugg-head { display: flex; align-items: center; justify-content: space-between; gap: 8px; padding: 2px 4px 6px; }
+.sugg-title { color: var(--muted); font-size: 10.5px; letter-spacing: .5px; text-transform: uppercase; }
+.sugg-list { display: flex; flex-direction: column; }
+.sugg-item { display: flex; align-items: baseline; gap: 8px; padding: 6px 9px; border-radius: 5px; cursor: pointer; }
+.sugg-item:hover { background: var(--amber-soft); }
+.sugg-empty { color: var(--muted); font-size: 11.5px; padding: 6px 9px; }
+.sugg-cat-head { display: flex; align-items: center; gap: 7px; padding: 7px 4px 2px; }
+.sugg-cat-head:first-child { padding-top: 2px; }
+.sugg-cat-name { color: var(--teal); font-family: var(--font-mono); font-size: 10px; letter-spacing: .6px; text-transform: uppercase; }
+.sugg-cat-count { color: var(--muted); font-size: 9.5px; font-family: var(--font-mono); background: var(--ink-3); border-radius: 8px; padding: 0 6px; }
+.sugg-more { margin-left: auto; font-size: 10px; padding: 0 4px; color: var(--muted); }
+
 /* Editor settings modal */
 .es-content { display: flex; flex-direction: column; gap: 18px; }
 .es-row { display: flex; flex-direction: column; gap: 5px; }
diff --git a/movie_editor/frontend/suggestions.js b/movie_editor/frontend/suggestions.js
new file mode 100644
index 00000000..3d651893
--- /dev/null
+++ b/movie_editor/frontend/suggestions.js
@@ -0,0 +1,223 @@
+// Shortcut suggestions ("Ideas" bulb): a quiet nudge next to prompt fields that lights
+// up when whole shortcut categories (camera moves, sub-actions, …) exist in the library
+// but are absent from the current prompt — easy to forget, and the video comes out plain.
+// Never opens on its own: click the bulb for a handful of insertable ideas, ⟳ rerolls.
+// Toggled by the "Shortcut ideas" editor setting. Reusable: ShortcutSuggest.bulb(textarea).
+(function () {
+  const { el } = window.dom;
+  const S = window.Store;
+
+  const MAX_CHIPS = 3;            // random samples shown per category (before "+N more")
+
+  function enabled() { return !!S.getEditorSetting("suggestions"); }
+
+  function libraryShortcuts() {
+    return (S.get().shortcuts || []).filter((s) => s.enabled !== false && (s.triggers || []).length);
+  }
+
+  // Does `text` contain `trigger` as a whole token (word boundaries on both sides)?
+  // Triggers may contain spaces, so this is a boundary-checked substring scan.
+  function containsTrigger(textLc, trigger) {
+    const t = trigger.toLowerCase();
+    if (!t) return false;
+    let at = textLc.indexOf(t);
+    const isWord = (ch) => /[a-z0-9_]/.test(ch || "");
+    while (at >= 0) {
+      const before = textLc[at - 1];
+      const after = textLc[at + t.length];
+      if (!isWord(before) && !isWord(after)) return true;
+      at = textLc.indexOf(t, at + 1);
+    }
+    return false;
+  }
+
+  // Group the library by category and split into covered (some trigger appears in the
+  // prompt) vs missing. The missing groups are the "you forgot about these" pool.
+  function analyze(text) {
+    const textLc = String(text || "").toLowerCase();
+    const groups = new Map();      // category -> { used: bool, items: [shortcut] }
+    for (const sc of libraryShortcuts()) {
+      const cat = sc.category || "other";
+      if (!groups.has(cat)) groups.set(cat, { used: false, items: [] });
+      const g = groups.get(cat);
+      g.items.push(sc);
+      if (!g.used && (sc.triggers || []).some((t) => containsTrigger(textLc, String(t || "").trim()))) {
+        g.used = true;
+      }
+    }
+    const missing = [...groups.entries()].filter(([, g]) => !g.used)
+      .map(([cat, g]) => ({ cat, items: g.items }));
+    return { missing, total: groups.size };
+  }
+
+  function shuffled(arr) {
+    const a = arr.slice();
+    for (let i = a.length - 1; i > 0; i--) {
+      const j = Math.floor(Math.random() * (i + 1));
+      [a[i], a[j]] = [a[j], a[i]];
+    }
+    return a;
+  }
+
+  // Insert at the caret (replacing any selection), padded with spaces against
+  // neighboring words, and fire `input` so the field's normal handlers persist it.
+  function insertIntoTextarea(ta, text) {
+    const v = ta.value;
+    let s = ta.selectionStart, e = ta.selectionEnd;
+    if (s == null) { s = e = v.length; }
+    const before = v.slice(0, s), after = v.slice(e);
+    const lead = before && !/\s$/.test(before) ? " " : "";
+    const trail = !/^(\s|[,;])/.test(after) ? " " : "";
+    const ins = lead + text + trail;
+    ta.value = before + ins + after;
+    const caret = s + ins.length;
+    ta.setSelectionRange(caret, caret);
+    ta.dispatchEvent(new Event("input", { bubbles: true }));
+    ta.focus();
+  }
+
+  function bulb(ta) {
+    const btn = el("button", "btn ghost tiny sugg-bulb", "💡");
+    btn.type = "button";
+    if (!enabled()) { btn.style.display = "none"; return btn; }
+
+    let pop = null;
+    let unlisten = null;           // detaches the document-level dismiss listeners
+    function closePop() {
+      if (pop) { pop.remove(); pop = null; }
+      if (unlisten) { unlisten(); unlisten = null; }
+    }
+
+    function refresh() {
+      const { missing } = analyze(ta.value);
+      const n = missing.length;
+      btn.classList.toggle("on", n > 0 && !!ta.value.trim());
+      btn.title = n > 0
+        ? `Ideas — ${n} shortcut ${n === 1 ? "category" : "categories"} this prompt isn't using`
+        : "Ideas — this prompt already touches every shortcut category";
+    }
+
+    function position() {
+      if (!pop) return;
+      const r = btn.getBoundingClientRect();
+      pop.style.top = (r.bottom + 4) + "px";
+      pop.style.left = Math.max(8, Math.min(r.left, window.innerWidth - 340)) + "px";
+    }
+
+    function chipRow(sc) {
+      const trig = String((sc.triggers || [])[0] || "").trim();
+      if (!trig) return null;
+      const row = el("div", "sugg-item");
+      row.append(el("span", "ac-trigger", trig));
+      const rep = (sc.replacements || [])[0];
+      if (rep) row.append(el("span", "ac-prompt", rep));
+      if (sc.sub_category) row.append(el("span", "ac-cat", sc.sub_category));
+      row.title = rep || trig;
+      // mousedown (not click) so the textarea keeps focus context, same as autocomplete.
+      row.addEventListener("mousedown", (ev) => {
+        ev.preventDefault();
+        insertIntoTextarea(ta, trig);
+        closePop();
+        refresh();
+      });
+      return row;
+    }
+
+    function openPop() {
+      closePop();
+      document.querySelectorAll(".sugg-pop").forEach((p) => p.remove());
+      // Never open off an invisible/detached bulb — the popover would strand at (0,0).
+      const br = btn.getBoundingClientRect();
+      if (!btn.isConnected || (!br.width && !br.height)) return;
+
+      const { missing } = analyze(ta.value);
+      pop = el("div", "sugg-pop");
+
+      const head = el("div", "sugg-head");
+      head.append(el("span", "sugg-title", missing.length
+        ? "Categories not in this prompt yet"
+        : "All categories covered ✓"));
+      const reroll = el("button", "btn ghost tiny", "⟳");
+      reroll.title = "Shuffle the samples";
+      reroll.onclick = (ev) => { ev.stopPropagation(); fill(); };
+      head.append(reroll);
+      pop.append(head);
+
+      const list = el("div", "sugg-list");
+      pop.append(list);
+
+      // Category-grouped browser: every missing category is listed (so you SEE what
+      // whole families you're skipping), with a few random samples under each and a
+      // "+N more" toggle to browse that category in full.
+      const expanded = new Set();
+      function fill() {
+        list.replaceChildren();
+        if (!missing.length) {
+          list.append(el("div", "sugg-empty",
+            "Nothing left to suggest — every shortcut category already appears here."));
+          return;
+        }
+        const cats = missing.slice().sort((a, b) => a.cat.localeCompare(b.cat));
+        for (const m of cats) {
+          const headRow = el("div", "sugg-cat-head");
+          headRow.append(el("span", "sugg-cat-name", m.cat));
+          headRow.append(el("span", "sugg-cat-count", String(m.items.length)));
+          const showAll = expanded.has(m.cat);
+          if (m.items.length > MAX_CHIPS) {
+            const more = el("button", "btn ghost tiny sugg-more",
+              showAll ? "less" : `+${m.items.length - MAX_CHIPS} more`);
+            more.onclick = (ev) => {
+              ev.stopPropagation();
+              showAll ? expanded.delete(m.cat) : expanded.add(m.cat);
+              fill();
+            };
+            headRow.append(more);
+          }
+          list.append(headRow);
+          const items = showAll ? m.items : shuffled(m.items).slice(0, MAX_CHIPS);
+          for (const sc of items) {
+            const row = chipRow(sc);
+            if (row) list.append(row);
+          }
+        }
+      }
+      fill();
+
+      document.body.append(pop);
+      position();
+      // Dismiss on any outside press or Escape; reposition while scrolling.
+      // Every close path funnels through closePop, which also detaches these.
+      const onDown = (ev) => { if (pop && !pop.contains(ev.target) && ev.target !== btn) closePop(); };
+      const onKey = (ev) => { if (ev.key === "Escape") closePop(); };
+      const onScroll = () => position();
+      document.addEventListener("mousedown", onDown, true);
+      document.addEventListener("keydown", onKey, true);
+      window.addEventListener("scroll", onScroll, true);
+      unlisten = () => {
+        document.removeEventListener("mousedown", onDown, true);
+        document.removeEventListener("keydown", onKey, true);
+        window.removeEventListener("scroll", onScroll, true);
+      };
+    }
+
+    btn.onclick = (ev) => { ev.preventDefault(); if (pop) closePop(); else openPop(); };
+
+    // Keep the glow honest while typing — cheap scan, debounced.
+    let timer = null;
+    ta.addEventListener("input", () => {
+      clearTimeout(timer);
+      timer = setTimeout(refresh, 400);
+    });
+    // Callers create the bulb before appending the field to the DOM — defer the first scan.
+    setTimeout(refresh, 0);
+    // A panel re-render can remove the bulb while its popover is open; close it (and
+    // stop watching) when the bulb detaches, same pattern as the autocomplete menu.
+    const detachObs = new MutationObserver(() => {
+      if (!btn.isConnected) { closePop(); detachObs.disconnect(); }
+    });
+    setTimeout(() => { if (btn.isConnected) detachObs.observe(document.body, { childList: true, subtree: true }); }, 0);
+    return btn;
+  }
+
+  window.ShortcutSuggest = { bulb };
+})();

From e163f4d0065259e3bb478b5445158bc0790a6a53 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 03:52:40 +0300
Subject: [PATCH 02/11] Media gallery: one-click continuity pin on image cards
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

📌 hover action on image cards sets/unsets continuity_settings.identity_pin_ref
(same setting as Engine settings ▸ Continuity ▸ Identity pin, just quicker).
The pinned card shows a persistent 📌 badge on its thumbnail. Pinning while
Auto continuity is explicitly off re-enables it — a pin that does nothing
would be a hidden setting. Image cards only, chain-sampler pipelines only.

fpMediabrowser now fingerprints identity_pin_ref so pin changes repaint the
gallery zone (they were dropped by the ViewBus fingerprint before).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 movie_editor/frontend/mediabrowser.js | 26 ++++++++++++++++++++++++++
 movie_editor/frontend/styles.css      |  5 +++++
 movie_editor/frontend/view_bus.js     |  2 ++
 3 files changed, 33 insertions(+)

diff --git a/movie_editor/frontend/mediabrowser.js b/movie_editor/frontend/mediabrowser.js
index eb9089f5..c7e7cd3b 100644
--- a/movie_editor/frontend/mediabrowser.js
+++ b/movie_editor/frontend/mediabrowser.js
@@ -306,6 +306,32 @@
       if (cur) _beginMediaRename(m, cur);
     };
     actions.append(ren);
+    // Continuity pin, right on the gallery card — same setting as Engine settings ▸
+    // Continuity ▸ Identity pin, just quicker to reach. Images only, chain-sampler
+    // pipelines only (custom pipelines have no auto-continuity machinery).
+    if (m.kind === "image" && st.project && window.PipelineCaps?.usesChainSampler(st)) {
+      const cs = st.project.continuity_settings || {};
+      const isPin = cs.identity_pin_ref === m.id;
+      const pinBtn = el("button", "media-act media-pin" + (isPin ? " active" : ""), "📌");
+      pinBtn.title = isPin
+        ? "Continuity pin — click to unpin"
+        : "Pin as identity guide for every scene (continuity pin)"
+          + (cs.auto_enabled === false ? " — also turns Auto continuity back on" : "");
+      pinBtn.onclick = (e) => {
+        e.stopPropagation();
+        const patch = { ...cs, identity_pin_ref: isPin ? null : m.id };
+        // A pin with auto continuity off does nothing — pinning means "use it", so
+        // re-enable (auto_enabled is on by default; only flip an explicit false).
+        if (!isPin && patch.auto_enabled === false) patch.auto_enabled = true;
+        S.patchProject({ continuity_settings: patch });
+      };
+      actions.append(pinBtn);
+      if (isPin) {
+        const badge = el("span", "media-pin-badge", "📌");
+        badge.title = "Continuity pin — identity guide for every scene";
+        thumb.append(badge);
+      }
+    }
     if (m.kind === "image" || m.kind === "video") {
       const exp = el("button", "media-act media-exp", "⤓");
       exp.title = "Export to disk";
diff --git a/movie_editor/frontend/styles.css b/movie_editor/frontend/styles.css
index 6dcc3da5..1250e6a0 100644
--- a/movie_editor/frontend/styles.css
+++ b/movie_editor/frontend/styles.css
@@ -321,6 +321,11 @@ input[type=number] { font-family: var(--font-mono); }
 }
 .media-act:hover { color: var(--amber-2); background: var(--ink-3); }
 .media-act.media-del:hover { color: var(--danger); }
+.media-act.media-pin { filter: grayscale(1); }
+.media-act.media-pin:hover { filter: none; }
+.media-act.media-pin.active { filter: none; background: var(--amber-soft); }
+.media-pin-badge { position: absolute; top: 3px; left: 3px; font-size: 10px; line-height: 1;
+  padding: 2px 3px; border-radius: 4px; background: rgba(0, 0, 0, .55); pointer-events: none; }
 .media-name-edit {
   flex: 1; min-width: 0; box-sizing: border-box; margin: 0; padding: 2px 4px; font-size: 10px;
   background: var(--ink-0); border: 1px solid var(--amber); border-radius: 4px; color: var(--text);
diff --git a/movie_editor/frontend/view_bus.js b/movie_editor/frontend/view_bus.js
index 3a02deb8..3768125a 100644
--- a/movie_editor/frontend/view_bus.js
+++ b/movie_editor/frontend/view_bus.js
@@ -129,6 +129,8 @@
       projects: st.projects?.length,
       media: st.mediaBin?.length,
       mediaPreview: st.mediaPreviewId,
+      // Continuity pin renders on gallery cards (📌 button state + thumb badge).
+      pin: st.project?.continuity_settings?.identity_pin_ref,
       shortcuts: st.shortcuts?.length,
       transitions: st.transitions?.length,
       nleEffects: st.nleEffects?.length,

From 8eb6f3704abeb964094c40b3c350d2c4a915e1fd Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 04:04:04 +0300
Subject: [PATCH 03/11] Ideas popover: habit-aware co-pilot for multi-shot
 prompt construction
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The bulb's popover now mixes what the editor knows about YOUR prompting habits
with library browsing, in four sections:
- You've paired these before — shortcuts you historically put in the same shot
  as what's in the prompt, ranked by co-occurrence (reason chip: with «X» ×N)
- Your usual next shot — what usually follows the previous scene's shortcuts
  in timeline order (scene inspector only; reason chip: after «X» ×N)
- Continue the thread — more from categories the prompt already uses
- New directions — the existing missing-category browser (counts, +N more)

Habits are mined on demand by GET /api/library/suggestion_stats: a text scan of
all saved projects' scene texts (triggers stay verbatim in scene texts, so no
new persistent state — project files remain the only source of truth). Pairs =
same scene, follows = consecutive scenes in timeline order; editorial cut
subclips and excluded scenes are skipped. One-off co-occurrences (<2) ignored.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 movie_editor/backend/bridge.py       |  64 ++++++++
 movie_editor/frontend/api.js         |   1 +
 movie_editor/frontend/inspector.js   |  19 ++-
 movie_editor/frontend/styles.css     |   7 +-
 movie_editor/frontend/suggestions.js | 218 +++++++++++++++++++++------
 movie_editor/server.py               |   9 ++
 6 files changed, 268 insertions(+), 50 deletions(-)

diff --git a/movie_editor/backend/bridge.py b/movie_editor/backend/bridge.py
index 8728e88f..249ed3a3 100644
--- a/movie_editor/backend/bridge.py
+++ b/movie_editor/backend/bridge.py
@@ -305,6 +305,70 @@ def delete_transition(name: str) -> dict:
     return {"transitions": ti()}
 
 
+def suggestion_stats() -> dict:
+    """Mine the user's OWN prompting habits from all saved projects: which shortcuts
+    land in the same scene (pairs) and which shot follows which (follows, in timeline
+    order). Scene texts keep triggers verbatim (editor prompt model), so this is a
+    plain text scan. Derived on demand from project files — no learned state stored.
+    Stats are keyed by shortcut `key` (the library's stable id)."""
+    import re
+
+    from . import projects as _projects
+
+    si, *_ = _library_fns()
+    pats: list[tuple[str, Any]] = []
+    for item in si():
+        if item.get("enabled") is False:
+            continue
+        key = str(item.get("key") or item.get("name") or "")
+        if not key:
+            continue
+        for trig in item.get("triggers") or []:
+            t = str(trig or "").strip().lower()
+            if t:
+                pats.append((key, re.compile(r"(?<![a-z0-9_])" + re.escape(t) + r"(?![a-z0-9_])")))
+    counts: dict[str, int] = {}
+    pairs: dict[tuple, int] = {}
+    follows: dict[tuple, int] = {}
+    scenes_scanned = 0
+    for meta in _projects.list_projects():
+        proj = _projects.get(meta.get("id", ""))
+        if proj is None:
+            continue
+        by_id = {s.id: s for s in proj.scenes}
+        order = [sid for sid in (proj.timeline_order or []) if sid in by_id]
+        if not order:
+            order = [s.id for s in proj.scenes]
+        prev: set[str] | None = None
+        for sid in order:
+            sc = by_id[sid]
+            # Editorial cuts share the root's text — only the root owns the prompt.
+            if sc.excluded or (sc.gen_unit_id and sc.cut_offset_frames):
+                continue
+            text = (sc.text or "").strip().lower()
+            if not text:
+                continue
+            present = {k for k, pat in pats if pat.search(text)}
+            scenes_scanned += 1
+            for k in present:
+                counts[k] = counts.get(k, 0) + 1
+            for a in present:
+                for b in present:
+                    if a < b:
+                        pairs[(a, b)] = pairs.get((a, b), 0) + 1
+            if prev:
+                for a in prev:
+                    for b in present:
+                        follows[(a, b)] = follows.get((a, b), 0) + 1
+            prev = present
+    return {
+        "scenes": scenes_scanned,
+        "counts": counts,
+        "pairs": [[a, b, n] for (a, b), n in pairs.items()],
+        "follows": [[a, b, n] for (a, b), n in follows.items()],
+    }
+
+
 def export_shortcuts() -> dict:
     return _funpack_attr("templates", "load_shortcut_db")()
 
diff --git a/movie_editor/frontend/api.js b/movie_editor/frontend/api.js
index 06946454..22be7d99 100644
--- a/movie_editor/frontend/api.js
+++ b/movie_editor/frontend/api.js
@@ -82,6 +82,7 @@
     importTransitions: (data, mode) => j("POST", API(`/library/transitions/import?mode=${mode || "merge"}`), data),
     clearTransitions: () => j("POST", API("/library/transitions/clear"), {}),
     shortcuts: () => j("GET", API("/library/shortcuts")),
+    suggestionStats: () => j("GET", API("/library/suggestion_stats")),
     saveCategory: (payload) => j("POST", API("/library/categories"), payload),
     saveShortcut: (item) => j("POST", API("/library/shortcuts"), item),
     deleteShortcut: (name) => j("DELETE", API(`/library/shortcuts/${encodeURIComponent(name)}`)),
diff --git a/movie_editor/frontend/inspector.js b/movie_editor/frontend/inspector.js
index f0711a0f..b2d04cf2 100644
--- a/movie_editor/frontend/inspector.js
+++ b/movie_editor/frontend/inspector.js
@@ -292,7 +292,24 @@
     // implicit control (double-fired clicks, stolen focus target).
     const promptWrap = el("div", "sugg-fieldwrap");
     promptWrap.append(field("Prompt", ta));
-    if (window.ShortcutSuggest) promptWrap.append(window.ShortcutSuggest.bulb(ta));
+    // prevText: the previous shot's prompt in TIMELINE order (skipping this scene's own
+    // editorial cuts and excluded scenes) — feeds the "usual next shot" suggestions.
+    const prevText = () => {
+      const p = S.get().project;
+      if (!p) return "";
+      const order = (Array.isArray(p.timeline_order) && p.timeline_order.length)
+        ? p.timeline_order : (p.scenes || []).map((s) => s.id);
+      const byId = new Map((p.scenes || []).map((s) => [s.id, s]));
+      const uid = S.genUnitId(scene);
+      for (let i = order.indexOf(scene.id) - 1; i >= 0; i--) {
+        const sc = byId.get(order[i]);
+        if (!sc || sc.excluded || S.genUnitId(sc) === uid) continue;
+        const prevRoot = S.genUnitRoot(S.genUnitId(sc)) || sc;
+        if ((prevRoot.text || "").trim()) return prevRoot.text;
+      }
+      return "";
+    };
+    if (window.ShortcutSuggest) promptWrap.append(window.ShortcutSuggest.bulb(ta, { prevText }));
     body.append(promptWrap);
     if (window.ShortcutAutocomplete) window.ShortcutAutocomplete.attach(ta);
     renderSourceField(st, root, scene);
diff --git a/movie_editor/frontend/styles.css b/movie_editor/frontend/styles.css
index 1250e6a0..c40a2e40 100644
--- a/movie_editor/frontend/styles.css
+++ b/movie_editor/frontend/styles.css
@@ -1571,7 +1571,7 @@ body.tour-mode { overflow: hidden; }
 .sugg-bulb.on { opacity: 1; filter: none; text-shadow: 0 0 7px rgba(255, 196, 87, .55); }
 .sugg-fieldwrap { position: relative; }
 .sugg-fieldwrap .sugg-bulb { position: absolute; top: -3px; right: 0; }
-.sugg-pop { position: fixed; z-index: 99999; width: 330px; max-height: 280px; overflow-y: auto;
+.sugg-pop { position: fixed; z-index: 99999; width: 330px; max-height: 360px; overflow-y: auto;
   background: #16140f; border: 1px solid var(--line); border-radius: 8px; box-shadow: var(--shadow); padding: 6px; }
 .sugg-head { display: flex; align-items: center; justify-content: space-between; gap: 8px; padding: 2px 4px 6px; }
 .sugg-title { color: var(--muted); font-size: 10.5px; letter-spacing: .5px; text-transform: uppercase; }
@@ -1579,6 +1579,11 @@ body.tour-mode { overflow: hidden; }
 .sugg-item { display: flex; align-items: baseline; gap: 8px; padding: 6px 9px; border-radius: 5px; cursor: pointer; }
 .sugg-item:hover { background: var(--amber-soft); }
 .sugg-empty { color: var(--muted); font-size: 11.5px; padding: 6px 9px; }
+.sugg-sec-head { color: var(--amber-2); font-size: 10px; letter-spacing: .6px; text-transform: uppercase;
+  padding: 8px 4px 3px; border-top: 1px solid var(--line); margin-top: 4px; }
+.sugg-habits:empty { display: none; }
+.sugg-habits .sugg-sec-head:first-child, .sugg-list > .sugg-sec-head:first-child { border-top: none; margin-top: 0; }
+.sugg-why { color: var(--muted); font-family: var(--font-mono); font-size: 9.5px; flex-shrink: 0; }
 .sugg-cat-head { display: flex; align-items: center; gap: 7px; padding: 7px 4px 2px; }
 .sugg-cat-head:first-child { padding-top: 2px; }
 .sugg-cat-name { color: var(--teal); font-family: var(--font-mono); font-size: 10px; letter-spacing: .6px; text-transform: uppercase; }
diff --git a/movie_editor/frontend/suggestions.js b/movie_editor/frontend/suggestions.js
index 3d651893..27b7c58e 100644
--- a/movie_editor/frontend/suggestions.js
+++ b/movie_editor/frontend/suggestions.js
@@ -1,13 +1,19 @@
-// Shortcut suggestions ("Ideas" bulb): a quiet nudge next to prompt fields that lights
-// up when whole shortcut categories (camera moves, sub-actions, …) exist in the library
-// but are absent from the current prompt — easy to forget, and the video comes out plain.
-// Never opens on its own: click the bulb for a handful of insertable ideas, ⟳ rerolls.
-// Toggled by the "Shortcut ideas" editor setting. Reusable: ShortcutSuggest.bulb(textarea).
+// Shortcut suggestions ("Ideas" bulb): a prompt-craft co-pilot next to prompt fields.
+// The bulb glows when whole shortcut categories go unused in the current prompt; the
+// popover mixes what it knows about YOUR habits (mined from all saved projects — pairs
+// used in the same shot, what shot usually follows the previous one) with browsing:
+// continue inside categories you're already using, or jump into ones you're not.
+// Never opens on its own. Toggled by the "Shortcut ideas" editor setting.
+// Reusable: ShortcutSuggest.bulb(textarea, { prevText: () => "…" }) — prevText supplies
+// the previous shot's prompt for "usual next shot" suggestions (scene inspector only).
 (function () {
   const { el } = window.dom;
   const S = window.Store;
+  const API = window.MovieEditorAPI;
 
   const MAX_CHIPS = 3;            // random samples shown per category (before "+N more")
+  const MAX_HABIT = 4;            // ranked rows per habit section (pairs / next shot)
+  const MIN_HABIT_N = 2;          // ignore one-off co-occurrences — not a habit yet
 
   function enabled() { return !!S.getEditorSetting("suggestions"); }
 
@@ -15,6 +21,8 @@
     return (S.get().shortcuts || []).filter((s) => s.enabled !== false && (s.triggers || []).length);
   }
 
+  function keyOf(sc) { return String(sc.key || sc.name || (sc.triggers || [])[0] || ""); }
+
   // Does `text` contain `trigger` as a whole token (word boundaries on both sides)?
   // Triggers may contain spaces, so this is a boundary-checked substring scan.
   function containsTrigger(textLc, trigger) {
@@ -31,6 +39,17 @@
     return false;
   }
 
+  function usesShortcut(textLc, sc) {
+    return (sc.triggers || []).some((t) => containsTrigger(textLc, String(t || "").trim()));
+  }
+
+  // Shortcuts whose trigger appears in `text`.
+  function presentIn(text) {
+    const lc = String(text || "").toLowerCase();
+    if (!lc.trim()) return [];
+    return libraryShortcuts().filter((sc) => usesShortcut(lc, sc));
+  }
+
   // Group the library by category and split into covered (some trigger appears in the
   // prompt) vs missing. The missing groups are the "you forgot about these" pool.
   function analyze(text) {
@@ -41,13 +60,13 @@
       if (!groups.has(cat)) groups.set(cat, { used: false, items: [] });
       const g = groups.get(cat);
       g.items.push(sc);
-      if (!g.used && (sc.triggers || []).some((t) => containsTrigger(textLc, String(t || "").trim()))) {
-        g.used = true;
-      }
+      if (!g.used && usesShortcut(textLc, sc)) g.used = true;
     }
     const missing = [...groups.entries()].filter(([, g]) => !g.used)
       .map(([cat, g]) => ({ cat, items: g.items }));
-    return { missing, total: groups.size };
+    const used = [...groups.entries()].filter(([, g]) => g.used)
+      .map(([cat, g]) => ({ cat, items: g.items }));
+    return { missing, used, total: groups.size };
   }
 
   function shuffled(arr) {
@@ -59,6 +78,51 @@
     return a;
   }
 
+  // ── habit stats (mined server-side from all saved projects) ────────────────────
+  // {counts, pairs:[[a,b,n]], follows:[[a,b,n]]} keyed by shortcut library key.
+  function fetchStats() {
+    if (!API || !API.suggestionStats) return Promise.resolve(null);
+    return API.suggestionStats().catch(() => null);
+  }
+
+  function statMaps(stats) {
+    const pair = new Map();        // key -> Map(partner -> n), symmetric
+    const addPair = (a, b, n) => {
+      if (!pair.has(a)) pair.set(a, new Map());
+      pair.get(a).set(b, (pair.get(a).get(b) || 0) + n);
+    };
+    (stats.pairs || []).forEach(([a, b, n]) => { addPair(a, b, n); addPair(b, a, n); });
+    const follow = new Map();      // a -> Map(b -> n), directional: shot a then shot b
+    (stats.follows || []).forEach(([a, b, n]) => {
+      if (!follow.has(a)) follow.set(a, new Map());
+      follow.get(a).set(b, (follow.get(a).get(b) || 0) + n);
+    });
+    return { pair, follow };
+  }
+
+  // Rank candidates by habit strength against a set of context keys. Returns
+  // [{sc, n, withKey}] — n = total co-occurrences, withKey = strongest partner.
+  function rankByHabit(map, contextKeys, excludeKeys) {
+    const byKey = new Map(libraryShortcuts().map((sc) => [keyOf(sc), sc]));
+    const score = new Map();       // candidate key -> { n, best, bestN }
+    for (const ctx of contextKeys) {
+      const partners = map.get(ctx);
+      if (!partners) continue;
+      for (const [cand, n] of partners) {
+        if (excludeKeys.has(cand) || !byKey.has(cand)) continue;
+        const cur = score.get(cand) || { n: 0, best: ctx, bestN: 0 };
+        cur.n += n;
+        if (n > cur.bestN) { cur.best = ctx; cur.bestN = n; }
+        score.set(cand, cur);
+      }
+    }
+    return [...score.entries()]
+      .filter(([, v]) => v.n >= MIN_HABIT_N)
+      .sort((a, b) => b[1].n - a[1].n)
+      .slice(0, MAX_HABIT)
+      .map(([cand, v]) => ({ sc: byKey.get(cand), n: v.n, withSc: byKey.get(v.best) }));
+  }
+
   // Insert at the caret (replacing any selection), padded with spaces against
   // neighboring words, and fire `input` so the field's normal handlers persist it.
   function insertIntoTextarea(ta, text) {
@@ -76,7 +140,8 @@
     ta.focus();
   }
 
-  function bulb(ta) {
+  function bulb(ta, opts) {
+    opts = opts || {};
     const btn = el("button", "btn ghost tiny sugg-bulb", "💡");
     btn.type = "button";
     if (!enabled()) { btn.style.display = "none"; return btn; }
@@ -94,7 +159,7 @@
       btn.classList.toggle("on", n > 0 && !!ta.value.trim());
       btn.title = n > 0
         ? `Ideas — ${n} shortcut ${n === 1 ? "category" : "categories"} this prompt isn't using`
-        : "Ideas — this prompt already touches every shortcut category";
+        : "Ideas — pairings, next-shot habits, and category browse";
     }
 
     function position() {
@@ -104,14 +169,15 @@
       pop.style.left = Math.max(8, Math.min(r.left, window.innerWidth - 340)) + "px";
     }
 
-    function chipRow(sc) {
+    function chipRow(sc, why) {
       const trig = String((sc.triggers || [])[0] || "").trim();
       if (!trig) return null;
       const row = el("div", "sugg-item");
       row.append(el("span", "ac-trigger", trig));
       const rep = (sc.replacements || [])[0];
       if (rep) row.append(el("span", "ac-prompt", rep));
-      if (sc.sub_category) row.append(el("span", "ac-cat", sc.sub_category));
+      if (why) row.append(el("span", "sugg-why", why));
+      else if (sc.sub_category) row.append(el("span", "ac-cat", sc.sub_category));
       row.title = rep || trig;
       // mousedown (not click) so the textarea keeps focus context, same as autocomplete.
       row.addEventListener("mousedown", (ev) => {
@@ -123,6 +189,12 @@
       return row;
     }
 
+    function secHead(box, label) {
+      box.append(el("div", "sugg-sec-head", label));
+    }
+
+    function trigOf(sc) { return String((sc.triggers || [])[0] || "").trim(); }
+
     function openPop() {
       closePop();
       document.querySelectorAll(".sugg-pop").forEach((p) => p.remove());
@@ -130,58 +202,108 @@
       const br = btn.getBoundingClientRect();
       if (!btn.isConnected || (!br.width && !br.height)) return;
 
-      const { missing } = analyze(ta.value);
+      const { missing, used } = analyze(ta.value);
+      const present = presentIn(ta.value);
+      const presentKeys = new Set(present.map(keyOf));
       pop = el("div", "sugg-pop");
 
       const head = el("div", "sugg-head");
-      head.append(el("span", "sugg-title", missing.length
-        ? "Categories not in this prompt yet"
-        : "All categories covered ✓"));
+      head.append(el("span", "sugg-title", "Ideas"));
       const reroll = el("button", "btn ghost tiny", "⟳");
       reroll.title = "Shuffle the samples";
-      reroll.onclick = (ev) => { ev.stopPropagation(); fill(); };
+      reroll.onclick = (ev) => { ev.stopPropagation(); fillBrowse(); };
       head.append(reroll);
       pop.append(head);
 
+      // ── habits (async): "you've paired these before" + "your usual next shot" ──
+      const habits = el("div", "sugg-habits");
+      pop.append(habits);
+      fetchStats().then((stats) => {
+        if (!pop || !stats) return;
+        const { pair, follow } = statMaps(stats);
+        // Pairings: shortcuts you historically put in the SAME shot as what's here.
+        if (presentKeys.size) {
+          const ideas = rankByHabit(pair, presentKeys, presentKeys);
+          if (ideas.length) {
+            secHead(habits, "You've paired these before");
+            for (const i of ideas) {
+              const row = chipRow(i.sc, `with «${trigOf(i.withSc)}» ×${i.n}`);
+              if (row) habits.append(row);
+            }
+          }
+        }
+        // Next shot: what usually comes after the PREVIOUS shot's shortcuts.
+        const prevText = typeof opts.prevText === "function" ? (opts.prevText() || "") : "";
+        const prevKeys = new Set(presentIn(prevText).map(keyOf));
+        if (prevKeys.size) {
+          const ideas = rankByHabit(follow, prevKeys, presentKeys);
+          if (ideas.length) {
+            secHead(habits, "Your usual next shot");
+            for (const i of ideas) {
+              const row = chipRow(i.sc, `after «${trigOf(i.withSc)}» ×${i.n}`);
+              if (row) habits.append(row);
+            }
+          }
+        }
+        position();
+      });
+
       const list = el("div", "sugg-list");
       pop.append(list);
 
-      // Category-grouped browser: every missing category is listed (so you SEE what
-      // whole families you're skipping), with a few random samples under each and a
-      // "+N more" toggle to browse that category in full.
+      // ── browse (sync): continue current categories + jump into missing ones ────
       const expanded = new Set();
-      function fill() {
+      function fillBrowse() {
         list.replaceChildren();
-        if (!missing.length) {
-          list.append(el("div", "sugg-empty",
-            "Nothing left to suggest — every shortcut category already appears here."));
-          return;
+        // Continue: more from categories this prompt already uses (minus what's in it).
+        const contCats = used
+          .map((u) => ({ cat: u.cat, items: u.items.filter((sc) => !presentKeys.has(keyOf(sc))) }))
+          .filter((u) => u.items.length);
+        if (contCats.length) {
+          secHead(list, "Continue the thread");
+          for (const u of shuffled(contCats).slice(0, 3)) {
+            const headRow = el("div", "sugg-cat-head");
+            headRow.append(el("span", "sugg-cat-name", "more " + u.cat));
+            list.append(headRow);
+            for (const sc of shuffled(u.items).slice(0, 2)) {
+              const row = chipRow(sc);
+              if (row) list.append(row);
+            }
+          }
         }
-        const cats = missing.slice().sort((a, b) => a.cat.localeCompare(b.cat));
-        for (const m of cats) {
-          const headRow = el("div", "sugg-cat-head");
-          headRow.append(el("span", "sugg-cat-name", m.cat));
-          headRow.append(el("span", "sugg-cat-count", String(m.items.length)));
-          const showAll = expanded.has(m.cat);
-          if (m.items.length > MAX_CHIPS) {
-            const more = el("button", "btn ghost tiny sugg-more",
-              showAll ? "less" : `+${m.items.length - MAX_CHIPS} more`);
-            more.onclick = (ev) => {
-              ev.stopPropagation();
-              showAll ? expanded.delete(m.cat) : expanded.add(m.cat);
-              fill();
-            };
-            headRow.append(more);
-          }
-          list.append(headRow);
-          const items = showAll ? m.items : shuffled(m.items).slice(0, MAX_CHIPS);
-          for (const sc of items) {
-            const row = chipRow(sc);
-            if (row) list.append(row);
+        // Jump: whole categories this prompt doesn't touch, with counts and browse.
+        if (missing.length) {
+          secHead(list, "New directions");
+          const cats = missing.slice().sort((a, b) => a.cat.localeCompare(b.cat));
+          for (const m of cats) {
+            const headRow = el("div", "sugg-cat-head");
+            headRow.append(el("span", "sugg-cat-name", m.cat));
+            headRow.append(el("span", "sugg-cat-count", String(m.items.length)));
+            const showAll = expanded.has(m.cat);
+            if (m.items.length > MAX_CHIPS) {
+              const more = el("button", "btn ghost tiny sugg-more",
+                showAll ? "less" : `+${m.items.length - MAX_CHIPS} more`);
+              more.onclick = (ev) => {
+                ev.stopPropagation();
+                showAll ? expanded.delete(m.cat) : expanded.add(m.cat);
+                fillBrowse();
+              };
+              headRow.append(more);
+            }
+            list.append(headRow);
+            const items = showAll ? m.items : shuffled(m.items).slice(0, MAX_CHIPS);
+            for (const sc of items) {
+              const row = chipRow(sc);
+              if (row) list.append(row);
+            }
           }
         }
+        if (!contCats.length && !missing.length) {
+          list.append(el("div", "sugg-empty",
+            "Every shortcut category already appears here — nothing left to browse."));
+        }
       }
-      fill();
+      fillBrowse();
 
       document.body.append(pop);
       position();
diff --git a/movie_editor/server.py b/movie_editor/server.py
index 54d5dda5..284f4544 100644
--- a/movie_editor/server.py
+++ b/movie_editor/server.py
@@ -1315,6 +1315,15 @@ if web is not None and PromptServer is not None:
         except Exception as e:  # noqa: BLE001
             return web.json_response({"shortcuts": [], "error": str(e)})
 
+    @routes.get(UI_PREFIX + "/api/library/suggestion_stats")
+    async def _suggestion_stats(_req):
+        """Shortcut usage mined from all saved projects (pairs / follows / counts) —
+        feeds the prompt Ideas popover. Empty stats on any error, never a 500."""
+        try:
+            return web.json_response(bridge.suggestion_stats())
+        except Exception as e:  # noqa: BLE001
+            return web.json_response({"scenes": 0, "counts": {}, "pairs": [], "follows": [], "error": str(e)})
+
     @routes.post(UI_PREFIX + "/api/library/shortcuts")
     async def _shortcut_save(req):
         try:

From 1112c2dcccc36db5e80a3e78b83b0b546d56dee0 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 05:11:33 +0300
Subject: [PATCH 04/11] =?UTF-8?q?ALG=20blur=20guides:=20standalone=20?=
 =?UTF-8?q?=E2=80=94=20guide/JoyAI=20tail=20blur=20no=20longer=20requires?=
 =?UTF-8?q?=20alg=5Fenabled?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

alg_blur_guides now works with the sampler's alg_enabled off: the anchor
stays sharp and only the newly-appended guide-attention frames (mid-scene
guide, carried guides, per-scene guides, JoyAI memory) are blurred, reusing
the same alg_strength / alg_sigma_threshold parameters and sigma window.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 samplers.py | 26 ++++++++++++++++----------
 1 file changed, 16 insertions(+), 10 deletions(-)

diff --git a/samplers.py b/samplers.py
index 5365d0cf..612ba0a3 100644
--- a/samplers.py
+++ b/samplers.py
@@ -1783,7 +1783,9 @@ def sample_funpack_distilled_flow(model, x, sigmas, extra_args=None, callback=No
       additionally blurs that many trailing video frames — the newly-appended guide-
       attention frames this scene (mid_scene_guide / carry_i2v_guides-as-guide /
       configured guides / JoyAI memory), set by the Scene Chain Sampler's alg_blur_guides
-      toggle right before this chunk's sample call. 0 = untouched (default).
+      toggle right before this chunk's sample call. 0 = untouched (default). Standalone:
+      works with alg_enabled off (anchor stays sharp, only the guide tail is blurred),
+      reusing the same alg_strength / alg_sigma_threshold parameters.
     - EXPERIMENTAL mg_enabled: Momentum Guidance (arXiv:2602.20360). Keeps a running EMA
       (decay=mg_decay) of the per-step ODE direction and blends the current step's
       direction toward it (weight=mg_strength) while sigma is BELOW mg_sigma_threshold —
@@ -1835,17 +1837,21 @@ def sample_funpack_distilled_flow(model, x, sigmas, extra_args=None, callback=No
     # denoise_mask) or if the packed-latent layout can't be read (helper returns None).
     # alg_guide_tail_frames (>0) extends the same idea, beyond the paper, to the trailing
     # guide-attention frames this scene appended (alg_blur_guides on the Scene Chain Sampler).
-    alg_sharp_latent_image = getattr(model, "latent_image", None) if alg_enabled else None
-    alg_active = bool(alg_enabled) and extra_args.get("denoise_mask") is not None and alg_sharp_latent_image is not None
+    # The two are independent: guide-tail blur runs standalone with alg_enabled off (anchor
+    # stays sharp), sharing the same alg_strength / alg_sigma_threshold blur parameters.
+    alg_guide_tail_frames = max(0, int(alg_guide_tail_frames))
+    alg_wanted = bool(alg_enabled) or alg_guide_tail_frames > 0
+    alg_sharp_latent_image = getattr(model, "latent_image", None) if alg_wanted else None
+    alg_active = alg_wanted and extra_args.get("denoise_mask") is not None and alg_sharp_latent_image is not None
     alg_blurred_latent_image = _alg_blur_frames(
         model, alg_sharp_latent_image, max(1.0, float(alg_strength)),
-        frame_indices=(0,), tail_count=int(alg_guide_tail_frames),
+        frame_indices=(0,) if alg_enabled else (), tail_count=alg_guide_tail_frames,
     ) if alg_active else None
     alg_active = alg_active and alg_blurred_latent_image is not None
     if alg_active:
-        print(f"[FunPack AV] ALG on (strength={alg_strength}, sigma_threshold={alg_sigma_threshold}, "
-              f"guide_tail_frames={alg_guide_tail_frames}) "
-              f"— anchor blurred while sigma > threshold")
+        print(f"[FunPack AV] ALG on (anchor={'on' if alg_enabled else 'off'}, strength={alg_strength}, "
+              f"sigma_threshold={alg_sigma_threshold}, guide_tail_frames={alg_guide_tail_frames}) "
+              f"— blurred while sigma > threshold")
 
     # EXPERIMENTAL Momentum Guidance (arXiv:2602.20360): EMA of the per-step ODE direction,
     # blended into the direction actually used once sigma drops below mg_sigma_threshold —
@@ -2060,11 +2066,11 @@ class FunPackDistilledFlowSampler:
                 }),
                 "alg_strength": ("FLOAT", {
                     "default": 2.0, "min": 1.0, "max": 4.0, "step": 0.1,
-                    "tooltip": "Downsample factor for the anchor blur (paper default 2.5, but 2.0 held character/i2v consistency noticeably better in testing here). Higher = blurrier anchor during the affected steps. Only used when alg_enabled.",
+                    "tooltip": "Downsample factor for the anchor blur (paper default 2.5, but 2.0 held character/i2v consistency noticeably better in testing here). Higher = blurrier anchor during the affected steps. Used when alg_enabled, and also by the Scene Chain Sampler's alg_blur_guides (which works without alg_enabled).",
                 }),
                 "alg_sigma_threshold": ("FLOAT", {
                     "default": 0.975, "min": 0.5, "max": 0.999, "step": 0.005,
-                    "tooltip": "Anchor stays blurred while sigma is above this value (the near-pure-noise steps), then swaps to sharp. Higher = narrower blurred window. Only used when alg_enabled.",
+                    "tooltip": "Anchor stays blurred while sigma is above this value (the near-pure-noise steps), then swaps to sharp. Higher = narrower blurred window. Used when alg_enabled, and also by the Scene Chain Sampler's alg_blur_guides (which works without alg_enabled).",
                 }),
                 "mg_enabled": ("BOOLEAN", {
                     "default": False,
@@ -2324,7 +2330,7 @@ class FunPackLTXAVSceneChainSampler:
                 }),
                 "alg_blur_guides": ("BOOLEAN", {
                     "default": False,
-                    "tooltip": "EXPERIMENTAL: extends ALG (see the sampler's alg_enabled) from just the i2v anchor to also blur newly-appended guide-attention frames this scene (mid_scene_guide / carry_i2v_guides-as-guide / configured per-scene guides / JoyAI memory), for the same early steps. No effect unless the chosen sampler has alg_enabled on, or no guide frames were appended this scene.",
+                    "tooltip": "EXPERIMENTAL: extends ALG (see the sampler's alg_enabled) from just the i2v anchor to also blur newly-appended guide-attention frames this scene (mid_scene_guide / carry_i2v_guides-as-guide / configured per-scene guides / JoyAI memory), for the same early steps. Standalone: works even with the sampler's alg_enabled off (anchor stays sharp), reusing its alg_strength / alg_sigma_threshold. Requires the FunPack Distilled Flow sampler; no effect if no guide frames were appended this scene.",
                 }),
                 "bounded_attention_enabled": ("BOOLEAN", {
                     "default": False,

From 6528acf504184f1b1ad6de91eb14ea709c8796cd Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 05:26:21 +0300
Subject: [PATCH 05/11] Shortcut revolver: no-repeat cycling through
 multi-replacement shortcuts
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Editor ▸ Settings ▸ Editor gains a 'Shortcut revolver' toggle (+ 'Random' sub-option).
When on, each shortcut with several replacements draws from a per-shortcut queue —
sequential (first…last, then restart) or shuffled once per cycle — and nothing repeats
until the whole set has fired. Off (default) keeps the seeded random pick unchanged.

- State is a sidecar (shortcut_revolver.json) next to the shortcut DB: settings +
  per-shortcut remaining queue + replacement-set fingerprint (editing a shortcut's
  replacements restarts its cycle; changing either setting resets all cycles).
- Generation commits the draw (refine_v2 positive prompt); every preview path
  (parse_timeline, split mapping, fragments) peeks the same stored queue without
  saving, so previews show exactly what the next Generate will use.
- Batch Training variants each commit a draw: N variants sweep N consecutive
  chambers instead of seed-rolled repeats.
- FunPackStudio.IS_CHANGED returns nan while the revolver is on so fixed-seed
  re-runs aren't served from ComfyUI's node cache (the next chamber must fire).

Verified live on local ComfyUI: settings round-trip, seed-independent stable peeks,
sequential cycle REP_ONE→TWO→THREE→ONE across processes, peek-after-commit shows the
next chamber. 9 new unit tests; full suite has the same 20 pre-existing failures.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 conditioning.py                          |  82 ++++++++++---
 movie_editor/backend/bridge.py           |  14 +++
 movie_editor/frontend/api.js             |   2 +
 movie_editor/frontend/editor_settings.js |  52 ++++++++-
 movie_editor/server.py                   |  14 +++
 templates.py                             | 129 +++++++++++++++++++--
 tests/test_shortcut_revolver.py          | 140 +++++++++++++++++++++++
 7 files changed, 411 insertions(+), 22 deletions(-)
 create mode 100644 tests/test_shortcut_revolver.py

diff --git a/conditioning.py b/conditioning.py
index d16165fb..3cf89fc4 100644
--- a/conditioning.py
+++ b/conditioning.py
@@ -663,11 +663,13 @@ def _expand_with_map(text):
     try:
         from .templates import (load_shortcut_db, shortcut_list, shortcut_key,
                                  _shortcut_replacements, _shortcut_trigger_pattern,
-                                 normalize_refinement_key)
+                                 normalize_refinement_key, load_revolver,
+                                 revolver_next_replacement)
     except ImportError:
         from templates import (load_shortcut_db, shortcut_list, shortcut_key,
                                _shortcut_replacements, _shortcut_trigger_pattern,
-                               normalize_refinement_key)
+                               normalize_refinement_key, load_revolver,
+                               revolver_next_replacement)
     import random
     from hashlib import md5
 
@@ -678,7 +680,7 @@ def _expand_with_map(text):
     except Exception:
         return text, whole
     candidates = []
-    for shortcut in (data.get("shortcuts", {}) or {}).values():
+    for db_key, shortcut in (data.get("shortcuts", {}) or {}).items():
         if not isinstance(shortcut, dict) or not bool(shortcut.get("enabled", True)):
             continue
         reps = _shortcut_replacements(shortcut.get("replacements", shortcut.get("replacement", [])))
@@ -689,14 +691,23 @@ def _expand_with_map(text):
         for trig in shortcut_list(shortcut.get("triggers", [])):
             pat = _shortcut_trigger_pattern(trig)
             if pat:
-                candidates.append((trig, pat, reps, sc_key, sc_name))
+                candidates.append((trig, pat, reps, sc_key, sc_name, str(db_key)))
     if not candidates:
         return text, whole
     candidates.sort(key=lambda it: len(shortcut_key(it[0])), reverse=True)
-    combined = "|".join(f"(?P<t{i}>{pat})" for i, (_, pat, _, _, _) in enumerate(candidates))
+    combined = "|".join(f"(?P<t{i}>{pat})" for i, (_, pat, _, _, _, _) in enumerate(candidates))
     if not combined:
         return text, whole
     rng = random.Random(int(md5(text.encode("utf-8")).hexdigest()[:12], 16))
+    # Shortcut revolver: peek the same stored cycle generation will draw from (never saved
+    # here) so split boundaries are found in the text generation will actually see.
+    try:
+        _revolver = load_revolver()
+    except Exception:
+        _revolver = {"enabled": False, "random": False, "state": {}}
+    _rev_on = bool(_revolver.get("enabled"))
+    _rev_state = _revolver.get("state") if isinstance(_revolver.get("state"), dict) else {}
+    _rev_random = bool(_revolver.get("random"))
     try:
         pattern = re.compile(combined, re.IGNORECASE | re.UNICODE)
     except re.error:
@@ -711,9 +722,13 @@ def _expand_with_map(text):
             parts.append(lit); exp_pos += len(lit)
         rep = None
         pc_key, pc_name = "", ""
-        for i, (_, _, reps, sc_key, sc_name) in enumerate(candidates):
+        for i, (_, _, reps, sc_key, sc_name, db_key) in enumerate(candidates):
             if m.group(f"t{i}") is not None:
-                rep = rng.choice(reps); pc_key, pc_name = sc_key, sc_name; break
+                if _rev_on and len(reps) > 1:
+                    rep = revolver_next_replacement(_rev_state, db_key, reps, _rev_random)
+                else:
+                    rep = rng.choice(reps)
+                pc_key, pc_name = sc_key, sc_name; break
         if rep is None:
             rep = m.group(0)
         pieces.append({"exp_start": exp_pos, "exp_end": exp_pos + len(rep),
@@ -6661,13 +6676,13 @@ class FunPackVideoRefinerV2(FunPackVideoRefiner):
     def _v2_prompt_key(self, prompt):
         return re.sub(r"\s+", " ", str(prompt or "").strip())
 
-    def _v2_apply_global_shortcuts(self, text, seed=0):
+    def _v2_apply_global_shortcuts(self, text, seed=0, revolver_commit=False):
         try:
             try:
                 from .templates import apply_prompt_shortcuts
             except ImportError:
                 from templates import apply_prompt_shortcuts
-            expanded, applied = apply_prompt_shortcuts(text, seed=seed)
+            expanded, applied = apply_prompt_shortcuts(text, seed=seed, revolver_commit=revolver_commit)
         except Exception as error:
             return str(text or ""), f"Shortcuts: unavailable ({error}).", []
         if not applied:
@@ -11729,8 +11744,13 @@ class FunPackVideoRefinerV2(FunPackVideoRefiner):
         # seeds to build N variant conditionings (Studio owns conditioning production).
         _raw_positive_prompt = str(positive_prompt or "")
         _raw_intent_prompt = str(user_intent_prompt or "")
-        positive_prompt, shortcut_status, shortcut_applied = self._v2_apply_global_shortcuts(positive_prompt, seed=seed)
+        # revolver_commit only on the real positive prompt: the one draw per run that advances
+        # the shortcut revolver's cycle. The intent prompt peeks, and does so FIRST — if it
+        # echoes the positive prompt's triggers, both then resolve to the same chamber (the
+        # same alignment the shared seed used to give them).
         user_intent_prompt, intent_shortcut_status, intent_shortcut_applied = self._v2_apply_global_shortcuts(user_intent_prompt, seed=seed)
+        positive_prompt, shortcut_status, shortcut_applied = self._v2_apply_global_shortcuts(
+            positive_prompt, seed=seed, revolver_commit=True)
         if intent_shortcut_applied:
             shortcut_status = (
                 f"{shortcut_status}\nIntent {intent_shortcut_status}"
@@ -12864,14 +12884,31 @@ class FunPackVideoRefinerV2(FunPackVideoRefiner):
             return (cond_i, m)
 
         try:
-            base_expansion = self._v2_apply_global_shortcuts(str(raw_prompt or ""), seed=seed)[0]
+            try:
+                from .templates import revolver_enabled as _revolver_enabled
+            except ImportError:
+                from templates import revolver_enabled as _revolver_enabled
+            revolver_on = bool(_revolver_enabled())
         except Exception:
-            base_expansion = str(raw_prompt or "")
+            revolver_on = False
         out = [tagged(base_entry, 0, base_prompt)]
-        cache = {base_expansion: base_entry}
+        if revolver_on:
+            # Shortcut revolver: each variant draw below COMMITS and advances the shared
+            # cycle, so N variants sweep N consecutive chambers (no repeats until the pool
+            # exhausts). Re-expanding the base here would both burn a chamber and collide
+            # with variant 1's draw — skip seeding the dedupe cache with it.
+            base_expansion = str(raw_prompt or "")
+            cache = {}
+        else:
+            try:
+                base_expansion = self._v2_apply_global_shortcuts(str(raw_prompt or ""), seed=seed)[0]
+            except Exception:
+                base_expansion = str(raw_prompt or "")
+            cache = {base_expansion: base_entry}
         for i in range(1, n):
             try:
-                exp_i = self._v2_apply_global_shortcuts(str(raw_prompt or ""), seed=seed + i)[0]
+                exp_i = self._v2_apply_global_shortcuts(str(raw_prompt or ""), seed=seed + i,
+                                                        revolver_commit=True)[0]
             except Exception:
                 exp_i = base_expansion
             entry = cache.get(exp_i)
@@ -14387,6 +14424,23 @@ class FunPackStudio:
             },
         }
 
+    @classmethod
+    def IS_CHANGED(cls, **kwargs):
+        # Shortcut revolver advances persistent per-shortcut state on every run, so a
+        # fixed-seed re-run must not be served from ComfyUI's cache while it's enabled —
+        # the next Generate is supposed to draw the next replacement. Normal input-hash
+        # caching applies when the revolver is off.
+        try:
+            try:
+                from .templates import revolver_enabled
+            except ImportError:
+                from templates import revolver_enabled
+            if revolver_enabled():
+                return float("nan")
+        except Exception:
+            pass
+        return ""
+
     @staticmethod
     def _is_output_connected(prompt, unique_id, output_index):
         return FunPackVideoRefiner()._is_output_connected(prompt, unique_id, output_index)
diff --git a/movie_editor/backend/bridge.py b/movie_editor/backend/bridge.py
index 249ed3a3..6c2f342b 100644
--- a/movie_editor/backend/bridge.py
+++ b/movie_editor/backend/bridge.py
@@ -285,6 +285,20 @@ def delete_shortcut(name: str) -> dict:
     return {"shortcuts": si(), "categories": _shortcut_categories()}
 
 
+def revolver_settings() -> dict:
+    """Shortcut-revolver settings (enabled + random order). State lives server-side next to
+    the shortcut DB so previews and generation share one cycle."""
+    return _funpack_attr("templates", "revolver_settings")()
+
+
+def set_revolver_settings(payload: dict) -> dict:
+    fn = _funpack_attr("templates", "set_revolver_settings")
+    return fn(
+        enabled=payload.get("enabled"),
+        random_order=payload.get("random"),
+    )
+
+
 def save_category(payload: dict) -> dict:
     """Add a category (and optionally a sub-category under it) to the managed list."""
     add = _funpack_attr("templates", "add_shortcut_category")
diff --git a/movie_editor/frontend/api.js b/movie_editor/frontend/api.js
index 22be7d99..054ae761 100644
--- a/movie_editor/frontend/api.js
+++ b/movie_editor/frontend/api.js
@@ -89,6 +89,8 @@
     exportShortcutsUrl: () => API("/library/shortcuts/export"),
     importShortcuts: (data, mode) => j("POST", API(`/library/shortcuts/import?mode=${mode || "merge"}`), data),
     clearShortcuts: () => j("POST", API("/library/shortcuts/clear"), {}),
+    revolverSettings: () => j("GET", API("/library/revolver")),
+    setRevolverSettings: (payload) => j("POST", API("/library/revolver"), payload),
 
     // FunPack file manager (Composer ▸ Files)
     listFiles: () => j("GET", API("/files")),
diff --git a/movie_editor/frontend/editor_settings.js b/movie_editor/frontend/editor_settings.js
index b77102a8..cdb72094 100644
--- a/movie_editor/frontend/editor_settings.js
+++ b/movie_editor/frontend/editor_settings.js
@@ -141,6 +141,55 @@
     return wrap;
   }
 
+  // Shortcut revolver: server-side library setting (not per-browser) — the no-repeat cycle
+  // is shared by previews and generation, so the toggle lives with the shortcut DB.
+  function revolverSection() {
+    const wrap = el("div", "es-section");
+    wrap.append(el("div", "es-section-title", "Shortcut revolver"));
+    wrap.append(el("div", "es-hint",
+      "For shortcuts with several replacements: remember which replacement fired and don’t reuse it "
+      + "until the WHOLE set has fired. Off = seeded random pick (may repeat). Saved with your shortcut "
+      + "library — applies to every browser and to generation."));
+
+    const main = el("label", "chk es-toggle");
+    const mainCb = el("input"); mainCb.type = "checkbox"; mainCb.disabled = true;
+    main.append(mainCb, el("span", null, "Shortcut revolver"));
+    wrap.append(main);
+
+    const sub = el("div", "es-subfields");
+    const rand = el("label", "chk es-toggle");
+    const randCb = el("input"); randCb.type = "checkbox"; randCb.disabled = true;
+    rand.append(randCb, el("span", null, "Random"));
+    sub.append(rand);
+    sub.append(el("div", "es-hint",
+      "Unchecked: replacements fire in order — first, second, … last, then the cycle restarts. "
+      + "Checked: any order, but still no repeats until the set is exhausted. Changing either "
+      + "option restarts all cycles."));
+    wrap.append(sub);
+
+    const apply = (s) => {
+      mainCb.checked = !!(s && s.enabled);
+      randCb.checked = !!(s && s.random);
+      mainCb.disabled = false;
+      randCb.disabled = !mainCb.checked;
+      sub.style.display = mainCb.checked ? "" : "none";
+    };
+    const save = () => {
+      mainCb.disabled = randCb.disabled = true;
+      API.setRevolverSettings({ enabled: mainCb.checked, random: randCb.checked })
+        .then(apply)
+        .catch(() => { API.revolverSettings().then(apply).catch(() => {}); });
+    };
+    mainCb.onchange = save;
+    randCb.onchange = save;
+
+    sub.style.display = "none";
+    API.revolverSettings().then(apply).catch(() => {
+      wrap.append(el("div", "es-hint", "Could not load revolver settings."));
+    });
+    return wrap;
+  }
+
   function toggleRow(label, hint, key) {
     const row = el("div", "es-row");
     const lbl = el("label", "chk es-toggle");
@@ -166,6 +215,7 @@
       "Use anchor",
       "Text before the first split trigger is a shared anchor prepended to every scene. Turn off to make that leading text Scene 1 instead.",
       "anchorEnabled"));
+    content.append(revolverSection());
     content.append(anchorGuideSection());
     body.append(content);
   }
@@ -173,7 +223,7 @@
   window.SettingsWindow.register({
     id: "editor", group: "", title: "Editor",
     subtitle: "Per-browser preferences — they apply to this browser, not the project.",
-    keywords: "autocomplete anchor prompt shortcuts ideas suggestions i2v bypass guide preferences",
+    keywords: "autocomplete anchor prompt shortcuts ideas suggestions i2v bypass guide preferences revolver random replacements cycle",
     iconBg: "linear-gradient(180deg,#6aa9ff,#3b6fd9)",
     icon: '<svg viewBox="0 0 16 16" width="13" height="13" fill="none" stroke="#fff" stroke-width="1.5" stroke-linecap="round"><path d="M2 4.5h12M2 8h12M2 11.5h12"/><circle cx="6" cy="4.5" r="1.7" fill="#fff" stroke="none"/><circle cx="11" cy="8" r="1.7" fill="#fff" stroke="none"/><circle cx="5" cy="11.5" r="1.7" fill="#fff" stroke="none"/></svg>',
     mount,
diff --git a/movie_editor/server.py b/movie_editor/server.py
index 284f4544..4d8c6f06 100644
--- a/movie_editor/server.py
+++ b/movie_editor/server.py
@@ -1338,6 +1338,20 @@ if web is not None and PromptServer is not None:
         except Exception as e:  # noqa: BLE001
             raise web.HTTPBadRequest(reason=str(e))
 
+    @routes.get(UI_PREFIX + "/api/library/revolver")
+    async def _revolver_get(_req):
+        try:
+            return web.json_response(bridge.revolver_settings())
+        except Exception as e:  # noqa: BLE001
+            return web.json_response({"enabled": False, "random": False, "error": str(e)})
+
+    @routes.post(UI_PREFIX + "/api/library/revolver")
+    async def _revolver_set(req):
+        try:
+            return web.json_response(bridge.set_revolver_settings(await req.json()))
+        except Exception as e:  # noqa: BLE001
+            raise web.HTTPBadRequest(reason=str(e))
+
     @routes.post(UI_PREFIX + "/api/library/categories")
     async def _category_save(req):
         try:
diff --git a/templates.py b/templates.py
index f7261147..6641d69c 100644
--- a/templates.py
+++ b/templates.py
@@ -548,6 +548,101 @@ def save_shortcut_db(data):
         json.dump(data, file, indent=2, sort_keys=True)
 
 
+REVOLVER_STORE_FILENAME = "shortcut_revolver.json"
+
+
+def revolver_store_path():
+    return os.path.join(template_store_dir(), REVOLVER_STORE_FILENAME)
+
+
+def load_revolver():
+    """Shortcut-revolver settings + per-shortcut cycle state. Sidecar of the shortcut DB —
+    deliberately NOT part of the shortcuts export/import payload. ``state[<shortcut key>]`` =
+    {"fp": <fingerprint of the replacement list>, "queue": [remaining replacement indices,
+    next one first]}."""
+    data = {}
+    path = revolver_store_path()
+    if os.path.exists(path):
+        try:
+            with open(path, "r", encoding="utf-8") as file:
+                data = json.load(file)
+        except (json.JSONDecodeError, OSError, ValueError):
+            data = {}
+    if not isinstance(data, dict):
+        data = {}
+    state = data.get("state")
+    return {
+        "enabled": bool(data.get("enabled", False)),
+        "random": bool(data.get("random", False)),
+        "state": state if isinstance(state, dict) else {},
+    }
+
+
+def save_revolver(data):
+    os.makedirs(template_store_dir(), exist_ok=True)
+    with open(revolver_store_path(), "w", encoding="utf-8") as file:
+        json.dump({
+            "enabled": bool(data.get("enabled", False)),
+            "random": bool(data.get("random", False)),
+            "state": data.get("state") if isinstance(data.get("state"), dict) else {},
+        }, file, indent=2)
+
+
+def revolver_enabled():
+    try:
+        return bool(load_revolver()["enabled"])
+    except Exception:
+        return False
+
+
+def revolver_settings():
+    data = load_revolver()
+    return {"enabled": data["enabled"], "random": data["random"]}
+
+
+def set_revolver_settings(enabled=None, random_order=None):
+    """Update revolver settings. Any actual change resets the cycle state — the old queues'
+    ordering semantics no longer apply once the mode flips."""
+    data = load_revolver()
+    changed = False
+    if enabled is not None and bool(enabled) != data["enabled"]:
+        data["enabled"] = bool(enabled)
+        changed = True
+    if random_order is not None and bool(random_order) != data["random"]:
+        data["random"] = bool(random_order)
+        changed = True
+    if changed:
+        data["state"] = {}
+        save_revolver(data)
+    return {"enabled": data["enabled"], "random": data["random"]}
+
+
+def _revolver_fingerprint(replacements):
+    payload = json.dumps(list(replacements), ensure_ascii=False)
+    return md5(payload.encode("utf-8")).hexdigest()[:12]
+
+
+def revolver_next_replacement(state, key, replacements, random_order):
+    """Draw the next replacement for shortcut ``key`` from its revolver queue, mutating
+    ``state`` in place. The queue is rebuilt — a fresh full cycle over every replacement —
+    when it's empty or the replacement set changed (fingerprint mismatch): sequential
+    (first, second, … last) by default, shuffled once per cycle when ``random_order``.
+    Either way nothing repeats until the whole set has fired."""
+    n = len(replacements)
+    fp = _revolver_fingerprint(replacements)
+    entry = state.get(key)
+    queue = None
+    if isinstance(entry, dict) and entry.get("fp") == fp and isinstance(entry.get("queue"), list):
+        queue = [int(i) for i in entry["queue"] if isinstance(i, int) and 0 <= int(i) < n]
+    if not queue:
+        queue = list(range(n))
+        if random_order:
+            random.shuffle(queue)
+    index = queue.pop(0)
+    state[key] = {"fp": fp, "queue": queue}
+    return replacements[index]
+
+
 def shortcut_items(data=None):
     data = load_shortcut_db() if data is None else normalize_shortcut_db(data)
     items = []
@@ -827,13 +922,18 @@ def resolve_variables(text, variables):
     return _expand(str(text or ""), frozenset(), 0), sorted(undefined)
 
 
-def apply_prompt_shortcuts(text, seed=0, shortcut_db=None):
+def apply_prompt_shortcuts(text, seed=0, shortcut_db=None, revolver_commit=False):
+    """Expand shortcut triggers. Multi-replacement picks are seeded-random unless the
+    shortcut revolver is enabled (see load_revolver): then each firing draws the next
+    replacement from the shortcut's no-repeat cycle. ``revolver_commit`` persists the
+    advanced cycle state — True only for real generation; previews peek without saving,
+    so they show exactly what the next generation will draw."""
     original = str(text or "")
     if not original:
         return original, []
     data = load_shortcut_db() if shortcut_db is None else normalize_shortcut_db(shortcut_db)
     candidates = []
-    for shortcut in data.get("shortcuts", {}).values():
+    for db_key, shortcut in data.get("shortcuts", {}).items():
         if not isinstance(shortcut, dict) or not bool(shortcut.get("enabled", True)):
             continue
         replacements = _shortcut_replacements(shortcut.get("replacements", shortcut.get("replacement", [])))
@@ -843,12 +943,13 @@ def apply_prompt_shortcuts(text, seed=0, shortcut_db=None):
         for trigger in shortcut_list(shortcut.get("triggers", [])):
             pattern = _shortcut_trigger_pattern(trigger)
             if pattern:
-                candidates.append((trigger, pattern, replacements, shortcut.get("name", trigger), sc_key))
+                candidates.append((trigger, pattern, replacements, shortcut.get("name", trigger),
+                                   sc_key, str(db_key)))
     if not candidates:
         return original, []
 
     candidates.sort(key=lambda item: len(shortcut_key(item[0])), reverse=True)
-    combined = "|".join(f"(?P<t{index}>{pattern})" for index, (_, pattern, _, _, _) in enumerate(candidates))
+    combined = "|".join(f"(?P<t{index}>{pattern})" for index, (_, pattern, _, _, _, _) in enumerate(candidates))
     if not combined:
         return original, []
     try:
@@ -858,15 +959,24 @@ def apply_prompt_shortcuts(text, seed=0, shortcut_db=None):
     if rng_seed == 0:
         rng_seed = int(md5(original.encode("utf-8")).hexdigest()[:12], 16)
     rng = random.Random(rng_seed)
+    revolver = load_revolver()
+    revolver_on = bool(revolver.get("enabled"))
+    revolver_state = revolver.get("state")
+    revolver_random = bool(revolver.get("random"))
+    revolver_dirty = False
     applied = []
     removals_happened = False
 
     def replace(match):
-        nonlocal removals_happened
-        for index, (trigger, _, replacements, name, sc_key) in enumerate(candidates):
+        nonlocal removals_happened, revolver_dirty
+        for index, (trigger, _, replacements, name, sc_key, db_key) in enumerate(candidates):
             if match.group(f"t{index}") is None:
                 continue
-            replacement = rng.choice(replacements)
+            if revolver_on and len(replacements) > 1:
+                replacement = revolver_next_replacement(revolver_state, db_key, replacements, revolver_random)
+                revolver_dirty = True
+            else:
+                replacement = rng.choice(replacements)
             applied.append({"name": str(name), "trigger": trigger,
                             "replacement": replacement, "refinement_key": sc_key})
             if not replacement:
@@ -877,6 +987,11 @@ def apply_prompt_shortcuts(text, seed=0, shortcut_db=None):
     expanded = re.sub(combined, replace, original, flags=re.IGNORECASE | re.UNICODE)
     if removals_happened:
         expanded = _cleanup_removed_phrases(expanded)
+    if revolver_dirty and revolver_commit:
+        try:
+            save_revolver(revolver)
+        except OSError:
+            pass
     return expanded, applied
 
 
diff --git a/tests/test_shortcut_revolver.py b/tests/test_shortcut_revolver.py
new file mode 100644
index 00000000..49cc891e
--- /dev/null
+++ b/tests/test_shortcut_revolver.py
@@ -0,0 +1,140 @@
+"""Shortcut revolver: no-repeat cycling through a shortcut's replacements.
+
+Covers: sequential cycle order + wrap-around, random mode as a full no-repeat permutation
+per cycle, peek (previews) vs commit (generation) semantics on apply_prompt_shortcuts,
+fingerprint reset when the replacement set is edited, settings changes resetting state,
+and off-by-default behavior staying byte-identical to the seeded random pick.
+"""
+import sys
+import types
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+
+class _FakeRoutes:
+    def get(self, _p):
+        return lambda f: f
+
+    def post(self, _p):
+        return lambda f: f
+
+    def put(self, _p):
+        return lambda f: f
+
+    def delete(self, _p):
+        return lambda f: f
+
+    def patch(self, _p):
+        return lambda f: f
+
+
+sys.modules.setdefault("folder_paths", types.SimpleNamespace(models_dir="", get_user_directory=lambda: ""))
+sys.modules.setdefault(
+    "server",
+    types.SimpleNamespace(PromptServer=types.SimpleNamespace(instance=types.SimpleNamespace(routes=_FakeRoutes()))),
+)
+
+import templates
+from templates import (
+    apply_prompt_shortcuts,
+    load_revolver,
+    normalize_shortcut_item,
+    revolver_next_replacement,
+    save_revolver,
+    set_revolver_settings,
+)
+
+REPS = ["alpha one", "alpha two", "alpha three"]
+
+
+def _db():
+    item = normalize_shortcut_item({
+        "name": "Alpha", "triggers": ["aa"], "replacements": list(REPS), "refinement_key": "",
+    })
+    single = normalize_shortcut_item({
+        "name": "Solo", "triggers": ["ss"], "replacements": ["only one"], "refinement_key": "",
+    })
+    return {"version": 1, "source": "test",
+            "shortcuts": {templates.shortcut_key("Alpha"): item,
+                          templates.shortcut_key("Solo"): single}}
+
+
+def _use_tmp_store(monkeypatch, tmp_path):
+    monkeypatch.setattr(templates, "template_store_dir", lambda: str(tmp_path))
+
+
+def test_sequential_cycle_then_wraps():
+    state = {}
+    draws = [revolver_next_replacement(state, "k", REPS, False) for _ in range(7)]
+    assert draws == REPS + REPS + [REPS[0]]
+
+
+def test_random_cycle_is_full_permutation_without_repeats():
+    state = {}
+    for _ in range(20):  # many cycles: each must be a permutation of the full set
+        cycle = [revolver_next_replacement(state, "k", REPS, True) for _ in range(len(REPS))]
+        assert sorted(cycle) == sorted(REPS)
+
+
+def test_replacement_edit_resets_cycle():
+    state = {}
+    revolver_next_replacement(state, "k", REPS, False)
+    edited = REPS + ["alpha four"]
+    assert revolver_next_replacement(state, "k", edited, False) == edited[0]
+
+
+def test_apply_shortcuts_peek_does_not_advance(monkeypatch, tmp_path):
+    _use_tmp_store(monkeypatch, tmp_path)
+    save_revolver({"enabled": True, "random": False, "state": {}})
+    first, _ = apply_prompt_shortcuts("an aa here", seed=1, shortcut_db=_db())
+    second, _ = apply_prompt_shortcuts("an aa here", seed=2, shortcut_db=_db())
+    assert first == second == f"an {REPS[0]} here"
+
+
+def test_apply_shortcuts_commit_advances_and_wraps(monkeypatch, tmp_path):
+    _use_tmp_store(monkeypatch, tmp_path)
+    save_revolver({"enabled": True, "random": False, "state": {}})
+    seen = [apply_prompt_shortcuts("an aa here", seed=1, shortcut_db=_db(),
+                                   revolver_commit=True)[0] for _ in range(4)]
+    assert seen == [f"an {r} here" for r in REPS + [REPS[0]]]
+
+
+def test_multiple_firings_in_one_prompt_consume_consecutive_chambers(monkeypatch, tmp_path):
+    _use_tmp_store(monkeypatch, tmp_path)
+    save_revolver({"enabled": True, "random": False, "state": {}})
+    expanded, applied = apply_prompt_shortcuts("aa then aa", seed=1, shortcut_db=_db(),
+                                               revolver_commit=True)
+    assert expanded == f"{REPS[0]} then {REPS[1]}"
+    assert [a["replacement"] for a in applied] == [REPS[0], REPS[1]]
+
+
+def test_single_replacement_shortcut_untouched_by_revolver(monkeypatch, tmp_path):
+    _use_tmp_store(monkeypatch, tmp_path)
+    save_revolver({"enabled": True, "random": False, "state": {}})
+    expanded, _ = apply_prompt_shortcuts("a ss here", seed=1, shortcut_db=_db(),
+                                         revolver_commit=True)
+    assert expanded == "a only one here"
+    assert load_revolver()["state"] == {}  # nothing multi-replacement fired -> no state written
+
+
+def test_disabled_matches_seeded_random(monkeypatch, tmp_path):
+    _use_tmp_store(monkeypatch, tmp_path)
+    a, _ = apply_prompt_shortcuts("an aa here", seed=7, shortcut_db=_db())
+    b, _ = apply_prompt_shortcuts("an aa here", seed=7, shortcut_db=_db(), revolver_commit=True)
+    assert a == b  # same seed -> same pick; no revolver file involved
+    assert not (tmp_path / templates.REVOLVER_STORE_FILENAME).exists()
+
+
+def test_settings_change_resets_state(monkeypatch, tmp_path):
+    _use_tmp_store(monkeypatch, tmp_path)
+    save_revolver({"enabled": True, "random": False,
+                   "state": {"x": {"fp": "abc", "queue": [2]}}})
+    settings = set_revolver_settings(random_order=True)
+    assert settings == {"enabled": True, "random": True}
+    assert load_revolver()["state"] == {}
+    # No-op update keeps state.
+    save_revolver({"enabled": True, "random": True,
+                   "state": {"x": {"fp": "abc", "queue": [2]}}})
+    set_revolver_settings(enabled=True, random_order=True)
+    assert load_revolver()["state"] != {}

From b8422d647b88eec02098339afc8ea7ed76391c5e Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 10:17:23 +0300
Subject: [PATCH 06/11] Guide blur: own strength + sigma-threshold controls,
 independent of ALG's
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

alg_blur_guides no longer borrows the sampler's anchor knobs: the Scene Chain
Sampler gains alg_guide_blur_strength (default 2.0) and
alg_guide_blur_sigma_threshold (default 0.975), threaded per-chunk via
extra_options like the tail count. The sampler now precomputes one latent per
(anchor blurred?, tail blurred?) combination — the frame sets are disjoint, so
the both-blurred variant composes the two passes — and each region follows its
own sigma window per step. Anchor ALG (alg_enabled/alg_strength/
alg_sigma_threshold) is untouched and back to anchor-only semantics.

Editor: 'Guide blur strength' and 'Guide blur sigma threshold' rows appear
under Engine ▸ Experimental when the blur toggle is on (sampler_inputs bridge,
names match the new node widgets; inputs appended at the block end to keep the
builder's positional widget mapping aligned).

Verified live: object_info shows both widgets appended last; Editor rows
appear on toggle and persist to project.sampler_inputs. 11 ALG unit tests
pass (new: composed independent-strength blur); same 20 pre-existing suite
failures.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 movie_editor/frontend/engine_settings.js |  4 +-
 samplers.py                              | 86 ++++++++++++++++++------
 tests/test_alg_guidance.py               | 33 +++++++++
 3 files changed, 103 insertions(+), 20 deletions(-)

diff --git a/movie_editor/frontend/engine_settings.js b/movie_editor/frontend/engine_settings.js
index 0b445b39..a88c9166 100644
--- a/movie_editor/frontend/engine_settings.js
+++ b/movie_editor/frontend/engine_settings.js
@@ -173,6 +173,8 @@
     { name: "joyai_audio_memory",    label: "Paired audio memory",   kind: "bool",  default: false, dependsOn: "joyai_memory" },
     { name: "v2a_grad_scale",        label: "Video→audio coupling", kind: "float", default: 1.0, min: 0.0, max: 4.0, step: 0.25, dependsOn: "joyai_audio_memory" },
     { name: "alg_blur_guides",       label: "Blur i2v guides and JoyAI memory", kind: "bool", default: false },
+    { name: "alg_guide_blur_strength", label: "Guide blur strength", kind: "float", default: 2.0, min: 1.0, max: 4.0, step: 0.1, dependsOn: "alg_blur_guides" },
+    { name: "alg_guide_blur_sigma_threshold", label: "Guide blur sigma threshold", kind: "float", default: 0.975, min: 0.5, max: 0.999, step: 0.005, dependsOn: "alg_blur_guides" },
     { name: "bounded_attention_enabled", label: "Bounded attention (multi-subject)", kind: "bool", default: false },
     { name: "dynashift",             label: "DynaShift (steer off bad gens)", kind: "bool", default: false },
     { name: "dynashift_strength",    label: "DynaShift strength",    kind: "float", default: 0.3, min: 0.05, max: 1.0, step: 0.05, dependsOn: "dynashift" },
@@ -281,7 +283,7 @@
     chain_timing: ["frame_overlap", "transition_duration", "use_same_seed"],
     chain_guidance: ["cfg", "embed_guidance", "embed_guidance_source", "embed_guidance_strength", "score_slider", "score_slider_strength", "output_guidance", "output_guidance_strength", "dynashift", "dynashift_strength", "dynashift_threshold"],
     chain_decode: ["decode_noise_scale", "decode_timestep", "decode_tile_size"],
-    chain_experimental: ["mid_scene_guide", "mid_scene_guide_strength", "joyai_memory", "joyai_memory_size", "joyai_fix_frames", "joyai_frame_select", "joyai_memory_strength", "joyai_audio_memory", "v2a_grad_scale", "alg_blur_guides", "bounded_attention_enabled"],
+    chain_experimental: ["mid_scene_guide", "mid_scene_guide_strength", "joyai_memory", "joyai_memory_size", "joyai_fix_frames", "joyai_frame_select", "joyai_memory_strength", "joyai_audio_memory", "v2a_grad_scale", "alg_blur_guides", "alg_guide_blur_strength", "alg_guide_blur_sigma_threshold", "bounded_attention_enabled"],
   };
 
   function countChainView(p, id) {
diff --git a/samplers.py b/samplers.py
index 612ba0a3..c91e304c 100644
--- a/samplers.py
+++ b/samplers.py
@@ -1755,6 +1755,7 @@ def sample_funpack_distilled_flow(model, x, sigmas, extra_args=None, callback=No
                                    rescue_prompt_sig=None,
                                    alg_enabled=False, alg_strength=2.0, alg_sigma_threshold=0.975,
                                    alg_guide_tail_frames=0,
+                                   alg_guide_blur_strength=2.0, alg_guide_blur_sigma_threshold=0.975,
                                    mg_enabled=False, mg_strength=0.5, mg_decay=0.5, mg_sigma_threshold=0.975,
                                    quality_sharpness=0.0):
     """
@@ -1784,8 +1785,10 @@ def sample_funpack_distilled_flow(model, x, sigmas, extra_args=None, callback=No
       attention frames this scene (mid_scene_guide / carry_i2v_guides-as-guide /
       configured guides / JoyAI memory), set by the Scene Chain Sampler's alg_blur_guides
       toggle right before this chunk's sample call. 0 = untouched (default). Standalone:
-      works with alg_enabled off (anchor stays sharp, only the guide tail is blurred),
-      reusing the same alg_strength / alg_sigma_threshold parameters.
+      works with alg_enabled off (anchor stays sharp, only the guide tail is blurred), with
+      its own independent controls — alg_guide_blur_strength / alg_guide_blur_sigma_threshold
+      (also set per-chunk by the Scene Chain Sampler), separate from the anchor's
+      alg_strength / alg_sigma_threshold.
     - EXPERIMENTAL mg_enabled: Momentum Guidance (arXiv:2602.20360). Keeps a running EMA
       (decay=mg_decay) of the per-step ODE direction and blends the current step's
       direction toward it (weight=mg_strength) while sigma is BELOW mg_sigma_threshold —
@@ -1837,20 +1840,43 @@ def sample_funpack_distilled_flow(model, x, sigmas, extra_args=None, callback=No
     # denoise_mask) or if the packed-latent layout can't be read (helper returns None).
     # alg_guide_tail_frames (>0) extends the same idea, beyond the paper, to the trailing
     # guide-attention frames this scene appended (alg_blur_guides on the Scene Chain Sampler).
-    # The two are independent: guide-tail blur runs standalone with alg_enabled off (anchor
-    # stays sharp), sharing the same alg_strength / alg_sigma_threshold blur parameters.
+    # Anchor and guide-tail blur are fully independent: each has its own strength and sigma
+    # window (alg_strength/alg_sigma_threshold vs alg_guide_blur_*), so one precomputed
+    # latent per (anchor blurred?, tail blurred?) combination that can occur — the frame
+    # sets are disjoint, so the both-blurred variant just composes the two.
     alg_guide_tail_frames = max(0, int(alg_guide_tail_frames))
-    alg_wanted = bool(alg_enabled) or alg_guide_tail_frames > 0
-    alg_sharp_latent_image = getattr(model, "latent_image", None) if alg_wanted else None
-    alg_active = alg_wanted and extra_args.get("denoise_mask") is not None and alg_sharp_latent_image is not None
-    alg_blurred_latent_image = _alg_blur_frames(
-        model, alg_sharp_latent_image, max(1.0, float(alg_strength)),
-        frame_indices=(0,) if alg_enabled else (), tail_count=alg_guide_tail_frames,
-    ) if alg_active else None
-    alg_active = alg_active and alg_blurred_latent_image is not None
+    alg_anchor_on = bool(alg_enabled)
+    alg_tail_on = alg_guide_tail_frames > 0
+    alg_sharp_latent_image = getattr(model, "latent_image", None) if (alg_anchor_on or alg_tail_on) else None
+    alg_active = ((alg_anchor_on or alg_tail_on) and extra_args.get("denoise_mask") is not None
+                  and alg_sharp_latent_image is not None)
+    alg_latents = {}
     if alg_active:
-        print(f"[FunPack AV] ALG on (anchor={'on' if alg_enabled else 'off'}, strength={alg_strength}, "
-              f"sigma_threshold={alg_sigma_threshold}, guide_tail_frames={alg_guide_tail_frames}) "
+        anchor_blurred = _alg_blur_frames(
+            model, alg_sharp_latent_image, max(1.0, float(alg_strength)), frame_indices=(0,),
+        ) if alg_anchor_on else None
+        tail_kappa = max(1.0, float(alg_guide_blur_strength))
+        tail_blurred = _alg_blur_frames(
+            model, alg_sharp_latent_image, tail_kappa, tail_count=alg_guide_tail_frames,
+        ) if alg_tail_on else None
+        both_blurred = _alg_blur_frames(
+            model, anchor_blurred, tail_kappa, tail_count=alg_guide_tail_frames,
+        ) if (anchor_blurred is not None and tail_blurred is not None) else None
+        alg_anchor_on = alg_anchor_on and anchor_blurred is not None
+        alg_tail_on = alg_tail_on and tail_blurred is not None
+        alg_latents = {
+            (False, False): alg_sharp_latent_image,
+            (True, False): anchor_blurred,
+            (False, True): tail_blurred,
+            (True, True): both_blurred if both_blurred is not None else (anchor_blurred or tail_blurred),
+        }
+        alg_active = alg_anchor_on or alg_tail_on
+    if alg_active:
+        anchor_desc = (f"strength={alg_strength}, sigma_threshold={alg_sigma_threshold}"
+                       if alg_anchor_on else "off")
+        tail_desc = (f"{alg_guide_tail_frames} frame(s), strength={alg_guide_blur_strength}, "
+                     f"sigma_threshold={alg_guide_blur_sigma_threshold}" if alg_tail_on else "off")
+        print(f"[FunPack AV] ALG on (anchor: {anchor_desc}; guide tail: {tail_desc}) "
               f"— blurred while sigma > threshold")
 
     # EXPERIMENTAL Momentum Guidance (arXiv:2602.20360): EMA of the per-step ODE direction,
@@ -1886,7 +1912,10 @@ def sample_funpack_distilled_flow(model, x, sigmas, extra_args=None, callback=No
             velocity_target = _velocity_bias_target(sigmas, sigma)
 
             if alg_active:
-                model.latent_image = alg_blurred_latent_image if float(sigma) > float(alg_sigma_threshold) else alg_sharp_latent_image
+                model.latent_image = alg_latents[(
+                    alg_anchor_on and float(sigma) > float(alg_sigma_threshold),
+                    alg_tail_on and float(sigma) > float(alg_guide_blur_sigma_threshold),
+                )]
 
             if _velocity_bias_enabled(velocity_bias_mode, "apply"):
                 x_pre = x
@@ -2066,11 +2095,11 @@ class FunPackDistilledFlowSampler:
                 }),
                 "alg_strength": ("FLOAT", {
                     "default": 2.0, "min": 1.0, "max": 4.0, "step": 0.1,
-                    "tooltip": "Downsample factor for the anchor blur (paper default 2.5, but 2.0 held character/i2v consistency noticeably better in testing here). Higher = blurrier anchor during the affected steps. Used when alg_enabled, and also by the Scene Chain Sampler's alg_blur_guides (which works without alg_enabled).",
+                    "tooltip": "Downsample factor for the anchor blur (paper default 2.5, but 2.0 held character/i2v consistency noticeably better in testing here). Higher = blurrier anchor during the affected steps. Only used when alg_enabled; guide-frame blur has its own controls on the Scene Chain Sampler (alg_blur_guides + alg_guide_blur_*).",
                 }),
                 "alg_sigma_threshold": ("FLOAT", {
                     "default": 0.975, "min": 0.5, "max": 0.999, "step": 0.005,
-                    "tooltip": "Anchor stays blurred while sigma is above this value (the near-pure-noise steps), then swaps to sharp. Higher = narrower blurred window. Used when alg_enabled, and also by the Scene Chain Sampler's alg_blur_guides (which works without alg_enabled).",
+                    "tooltip": "Anchor stays blurred while sigma is above this value (the near-pure-noise steps), then swaps to sharp. Higher = narrower blurred window. Only used when alg_enabled; guide-frame blur has its own controls on the Scene Chain Sampler (alg_blur_guides + alg_guide_blur_*).",
                 }),
                 "mg_enabled": ("BOOLEAN", {
                     "default": False,
@@ -2330,7 +2359,7 @@ class FunPackLTXAVSceneChainSampler:
                 }),
                 "alg_blur_guides": ("BOOLEAN", {
                     "default": False,
-                    "tooltip": "EXPERIMENTAL: extends ALG (see the sampler's alg_enabled) from just the i2v anchor to also blur newly-appended guide-attention frames this scene (mid_scene_guide / carry_i2v_guides-as-guide / configured per-scene guides / JoyAI memory), for the same early steps. Standalone: works even with the sampler's alg_enabled off (anchor stays sharp), reusing its alg_strength / alg_sigma_threshold. Requires the FunPack Distilled Flow sampler; no effect if no guide frames were appended this scene.",
+                    "tooltip": "EXPERIMENTAL: extends ALG (see the sampler's alg_enabled) from just the i2v anchor to also blur newly-appended guide-attention frames this scene (mid_scene_guide / carry_i2v_guides-as-guide / configured per-scene guides / JoyAI memory), for the same early steps. Standalone: works even with the sampler's alg_enabled off (anchor stays sharp), with its own alg_guide_blur_strength / alg_guide_blur_sigma_threshold controls below. Requires the FunPack Distilled Flow sampler; no effect if no guide frames were appended this scene.",
                 }),
                 "bounded_attention_enabled": ("BOOLEAN", {
                     "default": False,
@@ -2358,6 +2387,14 @@ class FunPackLTXAVSceneChainSampler:
                     "default": 0.6, "min": 0.3, "max": 0.95, "step": 0.05,
                     "tooltip": "Frame-similarity gate: a current frame must match a banked negative frame above this cosine similarity before any steering applies. Steering strength ramps from 0 at the threshold to full at similarity 1.0, so it self-releases once the unwanted feature is gone. Lower = more aggressive (risks steering away from legitimately similar content).",
                 }),
+                "alg_guide_blur_strength": ("FLOAT", {
+                    "default": 2.0, "min": 1.0, "max": 4.0, "step": 0.1,
+                    "tooltip": "Downsample factor for the guide-frame blur (alg_blur_guides). Higher = blurrier guide/JoyAI frames during the affected steps. Independent of the sampler's anchor alg_strength.",
+                }),
+                "alg_guide_blur_sigma_threshold": ("FLOAT", {
+                    "default": 0.975, "min": 0.5, "max": 0.999, "step": 0.005,
+                    "tooltip": "Guide frames stay blurred while sigma is above this value (the near-pure-noise steps), then swap to sharp. Higher = narrower blurred window. Independent of the sampler's anchor alg_sigma_threshold.",
+                }),
             },
             "hidden": {
                 "unique_id": "UNIQUE_ID",
@@ -2743,6 +2780,7 @@ class FunPackLTXAVSceneChainSampler:
 
     def _sample_chunk(self, model, sampler, sigmas, seed, cfg, positive, negative, latent,
                       pbar=None, step_offset=0, alg_guide_tail_frames=0,
+                      alg_guide_blur_strength=2.0, alg_guide_blur_sigma_threshold=0.975,
                       bounded_attention_enabled=False):
         if sampler is None:
             raise ValueError("sampler input is required.")
@@ -2765,6 +2803,8 @@ class FunPackLTXAVSceneChainSampler:
         extra_options = getattr(sampler, "extra_options", None)
         if isinstance(extra_options, dict) and sampler.sampler_function is sample_funpack_distilled_flow:
             extra_options["alg_guide_tail_frames"] = int(alg_guide_tail_frames)
+            extra_options["alg_guide_blur_strength"] = float(alg_guide_blur_strength)
+            extra_options["alg_guide_blur_sigma_threshold"] = float(alg_guide_blur_sigma_threshold)
 
         # EXPERIMENTAL Bounded Attention: model-level attention hooks (sampler-agnostic, unlike
         # the toggles above which only work on Distilled Flow), so install/remove here rather
@@ -4149,6 +4189,7 @@ class FunPackLTXAVSceneChainSampler:
                bounded_attention_enabled=False,
                output_guidance=False, output_guidance_strength=0.02,
                dynashift=False, dynashift_strength=0.3, dynashift_threshold=0.6,
+               alg_guide_blur_strength=2.0, alg_guide_blur_sigma_threshold=0.975,
                unique_id=None, prompt=None):
         if not isinstance(positive, list) or not positive:
             raise ValueError("positive conditioning must contain at least one scene entry.")
@@ -4200,6 +4241,8 @@ class FunPackLTXAVSceneChainSampler:
                 joyai_memory_strength=joyai_memory_strength,
                 joyai_audio_memory=joyai_audio_memory, v2a_grad_scale=v2a_grad_scale,
                 alg_blur_guides=alg_blur_guides,
+                alg_guide_blur_strength=alg_guide_blur_strength,
+                alg_guide_blur_sigma_threshold=alg_guide_blur_sigma_threshold,
                 bounded_attention_enabled=bounded_attention_enabled,
                 output_guidance=output_guidance, output_guidance_strength=output_guidance_strength,
                 dynashift=dynashift, dynashift_strength=dynashift_strength,
@@ -4482,6 +4525,8 @@ class FunPackLTXAVSceneChainSampler:
                     model, sampler, sigmas, scene_seed, cfg, scene_positive, scene_negative, chunk,
                     pbar=pbar, step_offset=scene_index * steps_per_scene,
                     alg_guide_tail_frames=(guide_tail if (alg_blur_guides and guide_tail > 0) else 0),
+                    alg_guide_blur_strength=alg_guide_blur_strength,
+                    alg_guide_blur_sigma_threshold=alg_guide_blur_sigma_threshold,
                     bounded_attention_enabled=bounded_attention_enabled,
                 )
                 _scene_sample_s = _time.perf_counter() - _t_sample0
@@ -4761,7 +4806,8 @@ class FunPackLTXAVSceneChainSampler:
                             joyai_audio_memory=False, v2a_grad_scale=1.0,
                             alg_blur_guides=False, bounded_attention_enabled=False,
                             output_guidance=False, output_guidance_strength=0.02,
-                            dynashift=False, dynashift_strength=0.3, dynashift_threshold=0.6):
+                            dynashift=False, dynashift_strength=0.3, dynashift_threshold=0.6,
+                            alg_guide_blur_strength=2.0, alg_guide_blur_sigma_threshold=0.975):
         """Sample one chain per Studio-packed variant entry (seed + index), persisting each result
         (latent + preview + per-entry cond + manifest) under ComfyUI temp for rating in Studio.
         Reuses sample() per entry with only the seed changed, so each entry is a clean generation."""
@@ -4808,6 +4854,8 @@ class FunPackLTXAVSceneChainSampler:
                 transition_duration=transition_duration,
                 decode_tile_size=decode_tile_size, refinement_key_input=key,
                 alg_blur_guides=alg_blur_guides,
+                alg_guide_blur_strength=alg_guide_blur_strength,
+                alg_guide_blur_sigma_threshold=alg_guide_blur_sigma_threshold,
                 bounded_attention_enabled=bounded_attention_enabled,
                 output_guidance=output_guidance, output_guidance_strength=output_guidance_strength,
                 dynashift=dynashift, dynashift_strength=dynashift_strength,
diff --git a/tests/test_alg_guidance.py b/tests/test_alg_guidance.py
index a6a50738..8b6a387b 100644
--- a/tests/test_alg_guidance.py
+++ b/tests/test_alg_guidance.py
@@ -137,6 +137,39 @@ def test_alg_blur_reduces_high_frequency_variance():
     assert float(frame0_out.var()) < float(frame0_in.var())
 
 
+def test_alg_blur_composes_with_independent_strengths():
+    """Anchor and guide-tail blur are separate passes with their own kappa: composing the
+    tail pass onto the anchor-blurred latent must leave the anchor frame untouched, and a
+    stronger tail kappa must blur the tail harder than the anchor kappa would."""
+    model = _fake_model()
+    latent_image = _packed_latent_image()
+    anchor_only = samplers._alg_blur_frames(model, latent_image, 2.0, frame_indices=(0,))
+    both = samplers._alg_blur_frames(model, anchor_only, 4.0, tail_count=1)
+    v_anchor_only = _video_frames(anchor_only)
+    v_both = _video_frames(both)
+
+    assert torch.equal(v_both[:, :, 0], v_anchor_only[:, :, 0])   # anchor pass preserved
+    assert torch.equal(v_both[:, :, 1:3], v_anchor_only[:, :, 1:3])  # middle untouched
+
+    # Strength independence needs content that survives downsampling distinctly (the 4x4
+    # checkerboard averages to uniform 0.5 at ANY factor) — use a seeded random 8x8 frame.
+    # Different kappa must yield a different blurred tail: proof the per-region strength
+    # actually reaches the filter (variance ordering is NOT guaranteed without antialias).
+    h8 = 8
+    model8 = _fake_model(shapes=((B, C, T, h8, h8), AUDIO_SHAPE))
+    video8 = torch.rand(B, C, T, h8, h8, generator=torch.Generator().manual_seed(7))
+    packed8 = torch.cat([video8.reshape(B, 1, C * T * h8 * h8),
+                         torch.full((B, 1, AUDIO_SIZE), 5.0)], dim=-1)
+    weak = samplers._alg_blur_frames(model8, packed8, 2.0, tail_count=1)
+    strong = samplers._alg_blur_frames(model8, packed8, 4.0, tail_count=1)
+    v_sz = C * T * h8 * h8
+    tail_weak = weak[..., :v_sz].reshape(B, C, T, h8, h8)[:, :, 3]
+    tail_strong = strong[..., :v_sz].reshape(B, C, T, h8, h8)[:, :, 3]
+    tail_orig = video8[:, :, 3]
+    assert not torch.allclose(tail_weak, tail_strong)  # kappa reaches the filter
+    assert not torch.allclose(tail_weak, tail_orig) and not torch.allclose(tail_strong, tail_orig)
+
+
 def test_alg_blur_is_deterministic():
     model = _fake_model()
     latent_image = _packed_latent_image()

From 0df109480c98c5d96897e153c663ec28a6c4b209 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 10:34:13 +0300
Subject: [PATCH 07/11] Preview stability: async segment encodes, ghost
 segments, cache-keyed trims, preload budget
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Four compounding defects behind stale/black previews on live instances:

1. /preview-segment ran its ffmpeg x264 re-encode synchronously inside the async
   handler — each encode froze ComfyUI's single aiohttp event loop for seconds,
   stalling every streaming /result response mid-transfer (elements error out,
   retries cache-bust and refetch, transport wedges). Worst right after a run,
   when the pool requests every clip's segment at once. Now asyncio.to_thread
   with a per-key lock (scrub storms don't spawn duplicate encodes) and
   -preset veryfast for preview trims. /export-clip wrapped the same way.

2. Ghosts were excluded from the segment system, so scrubbing a removed scene
   deep-seeked the raw moov-at-end chain MP4 — the exact stall segments were
   built to avoid. Ghosts (and lone clips with a deep in-point, e.g. the last
   surviving scene of a chain) now get segments; the trim window travels in the
   query (render_in + dur) since the scene no longer exists server-side.

3. Real staleness: segment URLs lacked the clip duration while responses are
   browser-cached for an hour — trimming a clip kept the same URL and replayed
   the hour-old segment. dur is now part of every segment URL.

4. Every pooled <video> used preload=auto: the whole timeline downloaded
   simultaneously, starving the browser's per-origin connections on tunneled
   instances and aborting the stream actually being watched. Pool now idles at
   preload=metadata; the current clip and its neighbours are promoted to auto.

Verified live on local ComfyUI with a synthetic 2-scene chain + ghost: ghost
segment 200/video-mp4, API stays responsive during an encode (17ms ping),
scrub across boundaries, seek into ghost (segment-relative time + badge),
play across the scene→ghost boundary, rapid pause/resume — all clean.
160 editor tests pass (new: ghost clip-spec parsing).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 movie_editor/frontend/api.js      |  4 ++
 movie_editor/frontend/player.js   | 53 +++++++++++++++--
 movie_editor/server.py            | 95 ++++++++++++++++++++++---------
 movie_editor/tests/test_server.py | 19 +++++++
 4 files changed, 141 insertions(+), 30 deletions(-)

diff --git a/movie_editor/frontend/api.js b/movie_editor/frontend/api.js
index 054ae761..f8788889 100644
--- a/movie_editor/frontend/api.js
+++ b/movie_editor/frontend/api.js
@@ -185,6 +185,10 @@
           type: m.type || "output",
           render_in: String(spec.renderIn != null ? spec.renderIn : 0),
         });
+        // dur is in the URL for two reasons: ghosts have no scene server-side (the trim
+        // window must travel in the query), and it versions the browser cache — segments
+        // are cached for an hour, so a timeline trim must produce a different URL.
+        if (spec.dur != null) q.set("dur", String(spec.dur));
         u += "?" + q.toString();
       }
       return u;
diff --git a/movie_editor/frontend/player.js b/movie_editor/frontend/player.js
index 30ab62c0..85f15b80 100644
--- a/movie_editor/frontend/player.js
+++ b/movie_editor/frontend/player.js
@@ -144,24 +144,38 @@
 
   // Multi-scene chain outputs share one ComfyUI file; deep seeks fail in the browser
   // unless the MP4 is faststart-trimmed like export. Serve per-scene segments instead.
+  // Ghosts (removed scenes still previewing their render) and lone clips with a deep
+  // in-point take the same path — they'd otherwise deep-seek the moov-at-end source
+  // directly, which stalls/blacks the monitor when scrubbing across them.
   function _assignChainPreviewSegments(st, clips) {
     const pid = st.project?.id;
     if (!pid || !API.previewSegmentUrl) return;
     const groups = new Map();
     for (const c of clips) {
-      if (c.pending || c.ghost || c.binUrl || !c.sceneId || !c.media) continue;
+      if (c.pending || c.binUrl || !c.media) continue;
+      if (!c.sceneId && !c.ghostId) continue;
       const k = _mediaChainKey(c);
       if (!k) continue;
       if (!groups.has(k)) groups.set(k, []);
       groups.get(k).push(c);
     }
     for (const list of groups.values()) {
-      if (list.length < 2) continue;
+      const shared = list.length >= 2;
       for (const c of list) {
+        if (!shared && (c.inSec || 0) <= 0.5) continue;  // shallow single clip: direct URL is fine
+        if (c.ghost) {
+          // No scene server-side — the trim window travels entirely in the query.
+          c.streamUrl = API.previewSegmentUrl(pid, c.ghostId, {
+            media: c.media, renderIn: c.inSec || 0, dur: c.durationSec,
+          });
+          continue;
+        }
         const sc = S.scene(c.sceneId);
         const sourceIn = sc?.source_in || 0;
         const renderIn = Math.max(0, (c.inSec || 0) - sourceIn);
-        c.streamUrl = API.previewSegmentUrl(pid, c.sceneId, { media: c.media, renderIn });
+        c.streamUrl = API.previewSegmentUrl(pid, c.sceneId, {
+          media: c.media, renderIn, dur: c.durationSec,
+        });
       }
     }
   }
@@ -474,7 +488,11 @@
     const zoom = document.createElement("div");
     zoom.className = "pm-fx-zoom";
     v = document.createElement("video");
-    v.className = "pm-video"; v.preload = "auto"; v.playsInline = true; v.src = url;
+    // preload starts as metadata: with preload=auto EVERY pooled clip downloaded at full
+    // tilt simultaneously — on a tunneled rental instance that starves the browser's
+    // per-origin connection pool and aborts the stream actually being watched.
+    // _updatePreloadHints() promotes the current clip and its neighbours to auto.
+    v.className = "pm-video"; v.preload = "metadata"; v.playsInline = true; v.src = url;
     v._pmFx = { viewport, zoom };
     zoom.append(v);
     viewport.append(zoom);
@@ -655,6 +673,29 @@
     else if (_badgeEl) _badgeEl.classList.remove("mixed-mode");
   }
 
+  // Promote the clip under the playhead and its neighbours to preload=auto (cut boundaries
+  // stay instant); everything else idles at metadata so the pool doesn't saturate the
+  // browser's per-origin connections. A hint change never interrupts an in-flight load.
+  function _updatePreloadHints() {
+    if (!pool.size) return;
+    const hot = new Set();
+    const add = (c) => {
+      if (!c || c.pending || c.blank) return;
+      const u = _clipUrl(c);
+      if (u) hot.add(u);
+    };
+    const idx = _currentClip ? _clips.indexOf(_currentClip) : -1;
+    if (idx >= 0) {
+      add(_clips[idx - 1]); add(_clips[idx]); add(_clips[idx + 1]);
+    } else {
+      add(_clips.find(_hasPlayableClipMedia));
+    }
+    for (const [u, v] of pool) {
+      const want = hot.has(u) ? "auto" : "metadata";
+      if (v.preload !== want) v.preload = want;
+    }
+  }
+
   function _sameVideoUrl(a, b) {
     if (!a || !b) return false;
     const ua = _clipUrl(a), ub = _clipUrl(b);
@@ -699,6 +740,7 @@
       else _clipBoundaryToken = 0;
     }
     _applyFx(clip, Math.max(0, offset));
+    _updatePreloadHints();
   }
 
   // Advance to the next clip at a clip's out-point (or stop). Same-source contiguous
@@ -728,6 +770,7 @@
         try { _applyFx(next, within); } catch (_) {}
       }
       if (_playing) _startTick();
+      _updatePreloadHints();
     }
     _notifyPh();
   }
@@ -820,6 +863,7 @@
       else _goto(clip, offset, false);
     }
     _syncInsAudio();
+    _updatePreloadHints();
     _notifyPh();
   }
 
@@ -1103,6 +1147,7 @@
         else if (_isBlankProgram()) _goto(_blankClip(), _phSec, _playing);
       }
     }
+    _updatePreloadHints();
 
     const p = st.project;
     const gen = st.gen || { state: "idle", media: [] };
diff --git a/movie_editor/server.py b/movie_editor/server.py
index 4d8c6f06..13835da0 100644
--- a/movie_editor/server.py
+++ b/movie_editor/server.py
@@ -184,7 +184,7 @@ def _clip_needs_trim(clip: dict) -> bool:
     return inn > 0.001 or dur is not None
 
 
-def _ffmpeg_trim_clip(src_path: str, out_path: str, inn, dur) -> None:
+def _ffmpeg_trim_clip(src_path: str, out_path: str, inn, dur, fast: bool = False) -> None:
     import shutil
     import subprocess
     ff = shutil.which("ffmpeg")
@@ -195,9 +195,12 @@ def _ffmpeg_trim_clip(src_path: str, out_path: str, inn, dur) -> None:
         cmd += ["-ss", f"{float(inn):.3f}"]
     if dur is not None:
         cmd += ["-t", f"{float(dur):.3f}"]
+    cmd += ["-i", src_path, "-c:v", "libx264", "-pix_fmt", "yuv420p"]
+    if fast:
+        # Preview segments: latency matters far more than a few % of bitrate — the player
+        # is waiting on this encode to unfreeze a scrub across a chain boundary.
+        cmd += ["-preset", "veryfast"]
     cmd += [
-        "-i", src_path,
-        "-c:v", "libx264", "-pix_fmt", "yuv420p",
         "-c:a", "aac", "-b:a", "192k",
         "-movflags", "+faststart", out_path,
     ]
@@ -221,6 +224,28 @@ def _playback_render_from_query(query) -> Optional[dict]:
     }
 
 
+def _ghost_playback_clip_spec(query) -> Optional[dict]:
+    """Preview-segment clip spec for a GHOST (removed scene whose render is still shown):
+    the scene no longer exists on the project, so the trim window comes entirely from the
+    query — in-point (render_in, absolute in the source file) and duration (dur)."""
+    filename = query.get("filename") if query else None
+    dur = query.get("dur") if query else None
+    if not filename or dur is None:
+        return None
+    try:
+        inn = float(query.get("render_in") or 0)
+        dur = float(dur)
+    except (TypeError, ValueError):
+        return None
+    return {
+        "filename": filename,
+        "subfolder": query.get("subfolder") or "",
+        "type": query.get("type") or "output",
+        "in": inn,
+        "dur": dur,
+    }
+
+
 def _scene_playback_clip_spec(
     project: Project,
     scene_id: str,
@@ -254,6 +279,7 @@ def _scene_playback_clip_spec(
 
 
 _preview_segment_cache: dict[str, str] = {}
+_preview_segment_locks: dict = {}  # cache key -> asyncio.Lock (dedupe concurrent encodes)
 
 
 def _preview_segment_cache_key(pid: str, scene_id: str, clip: dict) -> str:
@@ -1800,29 +1826,44 @@ if web is not None and PromptServer is not None:
         try:
             clip = _scene_playback_clip_spec(proj, scene_id, render_override=render_override)
         except KeyError as e:
-            return web.json_response({"detail": str(e)}, status=404)
+            # Ghosts: the scene was removed from the project but its render still previews;
+            # the trim window (in/dur) comes from the query instead of the scene.
+            clip = _ghost_playback_clip_spec(req.query)
+            if clip is None:
+                return web.json_response({"detail": str(e)}, status=404)
         key = _preview_segment_cache_key(pid, scene_id, clip)
-        cached = _preview_segment_cache.get(key)
-        if cached and os.path.isfile(cached):
-            return web.FileResponse(cached, headers={"Cache-Control": "private, max-age=3600"})
-        try:
-            import folder_paths
-            tempdir = folder_paths.get_temp_directory()
-        except Exception as e:  # noqa: BLE001
-            return web.json_response({"detail": f"Temp directory unavailable: {e}"}, status=500)
-        import time as _time
-        out_path = os.path.join(
-            tempdir,
-            f"funpack_preview_{pid[:8]}_{scene_id}_{int(_time.time() * 1000)}.mp4",
-        )
-        try:
-            src_path = _resolve_clip_src_path(clip)
-            _ffmpeg_trim_clip(src_path, out_path, clip.get("in"), clip.get("dur"))
-        except FileNotFoundError as e:
-            return web.json_response({"detail": str(e)}, status=400)
-        except RuntimeError as e:
-            return web.json_response({"detail": str(e)}, status=503)
-        _preview_segment_cache[key] = out_path
+        # Serialize per segment: the pool preloads every clip at once after a run, and a
+        # scrub can re-request the same URL — without this lock each request would start
+        # its own duplicate encode of the same segment.
+        lock = _preview_segment_locks.setdefault(key, asyncio.Lock())
+        async with lock:
+            cached = _preview_segment_cache.get(key)
+            if cached and os.path.isfile(cached):
+                return web.FileResponse(cached, headers={"Cache-Control": "private, max-age=3600"})
+            try:
+                import folder_paths
+                tempdir = folder_paths.get_temp_directory()
+            except Exception as e:  # noqa: BLE001
+                return web.json_response({"detail": f"Temp directory unavailable: {e}"}, status=500)
+            import time as _time
+            out_path = os.path.join(
+                tempdir,
+                f"funpack_preview_{pid[:8]}_{scene_id}_{int(_time.time() * 1000)}.mp4",
+            )
+            try:
+                src_path = _resolve_clip_src_path(clip)
+                # to_thread: this encode takes seconds — run inline it would freeze the ONE
+                # aiohttp event loop ComfyUI has, stalling every playing /result stream and
+                # the whole editor API for its duration (stale/black previews on scrub).
+                await asyncio.to_thread(
+                    _ffmpeg_trim_clip, src_path, out_path, clip.get("in"), clip.get("dur"),
+                    fast=True,
+                )
+            except FileNotFoundError as e:
+                return web.json_response({"detail": str(e)}, status=400)
+            except RuntimeError as e:
+                return web.json_response({"detail": str(e)}, status=503)
+            _preview_segment_cache[key] = out_path
         return web.FileResponse(out_path, headers={"Cache-Control": "private, max-age=3600"})
 
     # --- API: export one clip (ffmpeg trim from a chain output using in/dur) ---
@@ -1845,7 +1886,9 @@ if web is not None and PromptServer is not None:
         out_path = os.path.join(tempdir, out_name)
         try:
             src_path = _resolve_clip_src_path(clip)
-            _ffmpeg_trim_clip(src_path, out_path, clip.get("in"), clip.get("dur"))
+            # to_thread: same event-loop-freeze hazard as preview segments — an export
+            # during playback would stall every streaming /result response.
+            await asyncio.to_thread(_ffmpeg_trim_clip, src_path, out_path, clip.get("in"), clip.get("dur"))
         except FileNotFoundError as e:
             return web.json_response({"detail": str(e)}, status=400)
         except RuntimeError as e:
diff --git a/movie_editor/tests/test_server.py b/movie_editor/tests/test_server.py
index 251b6d30..88e239ae 100644
--- a/movie_editor/tests/test_server.py
+++ b/movie_editor/tests/test_server.py
@@ -421,6 +421,25 @@ def test_scene_playback_clip_spec_live_render_query():
     assert _playback_render_from_query({}) is None
 
 
+def test_ghost_playback_clip_spec_from_query():
+    """Ghost segments: the scene is gone from the project, so the trim window (in/dur)
+    must come entirely from the query — and malformed/incomplete queries return None
+    (the handler then 404s instead of encoding garbage)."""
+    from movie_editor.server import _ghost_playback_clip_spec
+    c = _ghost_playback_clip_spec({
+        "filename": "chain.mp4", "subfolder": "out", "type": "output",
+        "render_in": "7.4", "dur": "2.5",
+    })
+    assert c == {"filename": "chain.mp4", "subfolder": "out", "type": "output",
+                 "in": 7.4, "dur": 2.5}
+    assert _ghost_playback_clip_spec({"filename": "chain.mp4"}) is None          # no dur
+    assert _ghost_playback_clip_spec({"dur": "2.5"}) is None                      # no filename
+    assert _ghost_playback_clip_spec({"filename": "c.mp4", "dur": "oops"}) is None  # bad float
+    # render_in defaults to 0 (ghost of the chain's first scene).
+    c0 = _ghost_playback_clip_spec({"filename": "c.mp4", "dur": "4.0"})
+    assert c0["in"] == 0 and c0["type"] == "output"
+
+
 def test_project_models_keeps_empty_slots(monkeypatch):
     from movie_editor.backend import nodes
 

From 79b0b5bb1494fe79c0b3570418b9c7a0ee8fa6b4 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 20:08:26 +0300
Subject: [PATCH 08/11] Best-FaceID identity transfer: one-checkbox LoRA + LTX
 Identity Transfer splice
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Detects ComfyUI-BFSNodes' LTX Identity Transfer node in object_info and, when
enabled, auto-loads the Best-FaceID LoRA onto the model path and splices the
identity node between Studio and the Chain Sampler (project-wide, one
reference face per project — a run's graph can't vary it mid-chain). New
Engine settings pane under Chain Sampler picks the reference face (media
picker) and LoRA file (live dropdown), with advanced ArcFace/phase knobs.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---
 movie_editor/backend/builder.py          |  83 ++++++++++++
 movie_editor/backend/timeline.py         |   7 ++
 movie_editor/frontend/engine_settings.js | 153 +++++++++++++++++++++++
 movie_editor/server.py                   |  40 ++++++
 movie_editor/tests/test_builder.py       |  87 +++++++++++++
 5 files changed, 370 insertions(+)

diff --git a/movie_editor/backend/builder.py b/movie_editor/backend/builder.py
index 3e80f6f1..101f728f 100644
--- a/movie_editor/backend/builder.py
+++ b/movie_editor/backend/builder.py
@@ -93,6 +93,20 @@ CORE_PRODUCERS: list[tuple[str, int, str]] = [  # (core_id, output_index, type)
     ("f2i", 0, "INT"),
 ]
 
+# Best-FaceID identity transfer (ComfyUI-BFSNodes): a single project-wide splice, not a
+# generic Models-menu slot — it needs simultaneous model+positive+negative+vae+latent+
+# reference_face, a shape no ROLE covers. Detected by class presence so the feature
+# disappears cleanly when the custom node package isn't installed. Both names alias the
+# same node class (the package kept the old key as a back-compat alias).
+IDENTITY_TRANSFER_CLASSES: tuple[str, ...] = ("LTXIdentityTransfer", "LTXIdentityOverlapConditioning")
+
+
+def identity_transfer_class(object_info: dict) -> Optional[str]:
+    for cls in IDENTITY_TRANSFER_CLASSES:
+        if cls in (object_info or {}):
+            return cls
+    return None
+
 CONTROL_VALUES = {"fixed", "randomize", "increment", "decrement"}
 # Type-appropriate empty values for widgets that declare no default.
 _WIDGET_EMPTY = {"STRING": "", "INT": 0, "FLOAT": 0.0, "BOOLEAN": False}
@@ -542,6 +556,21 @@ def build(object_info: dict, models_config: dict, params: dict, media: dict | No
     # dropped here; an all-explicit loop is reported precisely. Never hand ComfyUI a cyclic graph.
     _break_cycles(graph, report, protected_edges, _node_labels(slots, slot_node_id, object_info))
 
+    # 6. Best-FaceID identity transfer: a fixed-core splice (not a slot), applied last so it
+    # can tap the already-resolved Studio · model and sampler · vae sources.
+    it_cfg = params.get("identity_transfer") or {}
+    if not disable_core and it_cfg.get("enabled"):
+        it_cls = identity_transfer_class(object_info)
+        if not it_cls:
+            msg = ("Identity Transfer (Best-FaceID) is enabled but no LTX Identity Transfer "
+                   "node is installed — install ComfyUI-BFSNodes.")
+            report["unsatisfied"].append(msg); report["blocking"].append(msg)
+        elif not it_cfg.get("reference_filename"):
+            msg = "Identity Transfer (Best-FaceID) is enabled but no reference face image is set."
+            report["unsatisfied"].append(msg); report["blocking"].append(msg)
+        else:
+            _splice_identity_transfer(graph, object_info, it_cfg, it_cls, report)
+
     for msg in pipeline_wiring.validate_models_wiring(models_config):
         report["blocking"].append(msg)
         report["unsatisfied"].append(msg)
@@ -906,3 +935,57 @@ def _apply_bypass(graph, slots, slot_node_id, slot_def, report):
                         del ndata["inputs"][inp_name]
         del graph[sid]
         report["wired"].append(f"{s.get('node_class')} bypassed (pass-through)")
+
+
+def _splice_identity_transfer(
+    graph: dict, object_info: dict, cfg: dict, cls: str, report: dict[str, list],
+) -> None:
+    """Insert the Best-FaceID LoRA + LTX Identity Transfer node between the fixed core's
+    loader->Studio edge and Studio/cond->sampler/concat edges.
+
+    Runs after every other wiring step, so it taps the already-resolved Studio · model and
+    sampler · vae sources and always wins over whatever fed sampler.model/positive/negative
+    and concat.video_latent — Identity Transfer isn't meant to be combined with manually
+    rewiring those exact ports in Full control.
+    """
+    lora_name = cfg.get("lora_name")
+    model_src = graph["studio"]["inputs"].get("model")
+    if lora_name:
+        lora_cls = "LoraLoaderModelOnly"
+        if lora_cls not in object_info:
+            msg = (f"Identity Transfer needs '{lora_cls}' (built into ComfyUI) but it's "
+                   "missing from the node registry.")
+            report["unsatisfied"].append(msg); report["blocking"].append(msg)
+        else:
+            lora_inputs = _widget_defaults(object_info.get(lora_cls))
+            lora_inputs["model"] = model_src
+            lora_inputs["lora_name"] = lora_name
+            lora_inputs["strength_model"] = float(cfg.get("lora_strength", 1.0))
+            graph["id_lora"] = {"class_type": lora_cls, "inputs": lora_inputs}
+            graph["studio"]["inputs"]["model"] = ["id_lora", 0]
+            report["wired"].append(f"Best-FaceID LoRA ({lora_name}) -> Studio · model")
+
+    graph["id_ref_load"] = {
+        "class_type": "LoadImage",
+        "inputs": {"image": cfg["reference_filename"]},
+    }
+
+    inputs = _widget_defaults(object_info.get(cls))
+    inputs.update({
+        "model": ["studio", 0],
+        "positive": ["cond", 0],
+        "negative": ["cond", 1],
+        "vae": graph["sampler"]["inputs"].get("vae"),
+        "latent": ["studio", 12],
+        "reference_face": ["id_ref_load", 0],
+    })
+    for k in ("identity_projector", "source_id", "phase_scale", "id_strength", "arcface_mode", "debug_log"):
+        if cfg.get(k) is not None:
+            inputs[k] = cfg[k]
+    graph["idxfer"] = {"class_type": cls, "inputs": inputs}
+
+    graph["sampler"]["inputs"]["model"] = ["idxfer", 0]
+    graph["sampler"]["inputs"]["positive"] = ["idxfer", 1]
+    graph["sampler"]["inputs"]["negative"] = ["idxfer", 2]
+    graph["concat"]["inputs"]["video_latent"] = ["idxfer", 3]
+    report["wired"].append("LTX Identity Transfer spliced between Studio and the sampler")
diff --git a/movie_editor/backend/timeline.py b/movie_editor/backend/timeline.py
index 34d20623..a55d859b 100644
--- a/movie_editor/backend/timeline.py
+++ b/movie_editor/backend/timeline.py
@@ -595,6 +595,12 @@ class Project:
     # (same shape as the global models.json). Empty {"slots": []} falls back to the
     # global default at build/read time; the editor seeds new projects from it.
     models: dict = field(default_factory=lambda: {"slots": []})
+    # Best-FaceID identity transfer (ComfyUI-BFSNodes' LTX Identity Transfer node), a single
+    # project-wide splice: {enabled, reference_media_ref, lora_name, lora_strength,
+    # identity_projector, source_id, phase_scale, id_strength, arcface_mode, debug_log}.
+    # One reference face for the whole project — the chain sampler runs multi-scene prompts
+    # through one graph per run, so there's no per-scene seam to hang a second identity on.
+    identity_transfer: dict = field(default_factory=dict)
     # Guide stack toggles — empty / stack_enabled=false keeps Studio carry behaviour.
     guide_settings: dict = field(default_factory=dict)
     continuity_settings: dict = field(default_factory=dict)
@@ -644,6 +650,7 @@ class Project:
             overlay_lanes=list(d.get("overlay_lanes") or []),
             overlay_tracks=list(d.get("overlay_tracks") or []),
             models=dict(d.get("models") or {"slots": []}),
+            identity_transfer=dict(d.get("identity_transfer") or {}),
             guide_settings=dict(d.get("guide_settings") or {}),
             continuity_settings=dict(d.get("continuity_settings") or {}),
             generation_meta=dict(d.get("generation_meta") or {}),
diff --git a/movie_editor/frontend/engine_settings.js b/movie_editor/frontend/engine_settings.js
index a88c9166..2ff2e221 100644
--- a/movie_editor/frontend/engine_settings.js
+++ b/movie_editor/frontend/engine_settings.js
@@ -5,6 +5,7 @@
 (function () {
   const { el, clear } = window.dom;
   const S = window.Store;
+  const API = window.MovieEditorAPI;
 
   let _mounted = null; // { scroller (pane, set per render), content (shell root) }
   let unsub = null;
@@ -214,6 +215,155 @@
     S.patchProject({ continuity_settings: { ...(p.continuity_settings || {}), ...patch } });
   }
 
+  // ── Best-FaceID identity transfer (ComfyUI-BFSNodes LTX Identity Transfer) ────────────
+  const IDENTITY_TRANSFER_CLASSES = ["LTXIdentityTransfer", "LTXIdentityOverlapConditioning"];
+  let idNodesInfo = null; // { available, loraChoices } — fetched lazily, cached for the session
+
+  async function ensureIdentityInfo(force) {
+    if (idNodesInfo && !force) return idNodesInfo;
+    const info = { available: false, loraChoices: [] };
+    try {
+      const nodes = (await API.allNodes()).nodes || [];
+      info.available = nodes.some((n) => IDENTITY_TRANSFER_CLASSES.includes(n.class));
+      if (info.available) {
+        const lora = await API.nodeSpec("LoraLoaderModelOnly");
+        const w = (lora?.inputs || []).find((i) => i.name === "lora_name");
+        info.loraChoices = (w && w.choices) || [];
+      }
+    } catch (_) { /* leave unavailable */ }
+    idNodesInfo = info;
+    render();
+    return info;
+  }
+
+  function normIdentityTransfer(p) {
+    const it = (p && p.identity_transfer) || {};
+    return {
+      enabled: !!it.enabled,
+      reference_media_ref: it.reference_media_ref || null,
+      lora_name: it.lora_name || "",
+      lora_strength: it.lora_strength != null ? +it.lora_strength : 1.0,
+      identity_projector: it.identity_projector || "None",
+      source_id: it.source_id != null ? +it.source_id : 2.0,
+      phase_scale: it.phase_scale != null ? +it.phase_scale : 1.0,
+      id_strength: it.id_strength != null ? +it.id_strength : 1.0,
+      arcface_mode: it.arcface_mode || "auto_adjust",
+      debug_log: !!it.debug_log,
+    };
+  }
+
+  function patchIdentityTransfer(patch) {
+    const p = S.get().project;
+    S.patchProject({ identity_transfer: { ...(p.identity_transfer || {}), ...patch } });
+  }
+
+  function renderChainIdentity(pane, st) {
+    const p = st.project;
+    const it = normIdentityTransfer(p);
+    const info = idNodesInfo;
+    if (!info) {
+      pane.append(hintEl("Checking for ComfyUI-BFSNodes…"));
+      return;
+    }
+    if (!info.available) {
+      pane.append(hintEl(
+        "Best-FaceID identity transfer needs the LTX Identity Transfer node from "
+        + "ComfyUI-BFSNodes (github.com/alisson-anjos/ComfyUI-BFSNodes) — install it in "
+        + "custom_nodes and restart ComfyUI."));
+      const retry = el("button", "btn ghost tiny", "Check again");
+      retry.type = "button";
+      retry.onclick = () => ensureIdentityInfo(true);
+      pane.append(retry);
+      return;
+    }
+    pane.append(hintEl(
+      "One reference face for the whole project. Loads the Best-FaceID LoRA on the model "
+      + "path and feeds the reference image into LTX Identity Transfer, spliced between "
+      + "Studio and the Chain Sampler. Tip: prefix your prompt with “ref_t2v:” for "
+      + "best results."));
+
+    const g = group(pane, "Best-FaceID");
+    const enCb = el("input"); enCb.type = "checkbox"; enCb.checked = it.enabled;
+    enCb.style.width = "auto";
+    enCb.dataset.k = "it-enabled";
+    enCb.onchange = () => patchIdentityTransfer({ enabled: enCb.checked });
+    g.append(toggleField("Enable identity transfer", enCb));
+
+    const refRow = el("div", "sw-row eng-field eng-stack");
+    const refMain = el("div", "sw-row-main");
+    refMain.append(el("div", "sw-row-title", "Reference face"));
+    refRow.append(refMain);
+    const refPicker = window.MediaPicker.create({
+      value: it.reference_media_ref,
+      mediaBin: st.mediaBin,
+      noneLabel: "— pick a reference face —",
+      onChange: (v) => patchIdentityTransfer({ reference_media_ref: v }),
+    });
+    if (!it.enabled) refPicker.classList.add("disabled");
+    refRow.append(refPicker);
+    g.append(refRow);
+
+    const loraChoices = info.loraChoices || [];
+    const loraSel = el("select"); loraSel.dataset.k = "it-lora";
+    loraSel.disabled = !it.enabled;
+    const noneOpt = el("option", null, "— select LoRA —"); noneOpt.value = "";
+    loraSel.append(noneOpt);
+    loraChoices.forEach((c) => {
+      const o = el("option", null, c); o.value = c;
+      if (c === it.lora_name) o.selected = true;
+      loraSel.append(o);
+    });
+    loraSel.onchange = () => patchIdentityTransfer({ lora_name: loraSel.value });
+    g.append(field("Best-FaceID LoRA", loraSel));
+
+    const strIn = el("input"); strIn.type = "number";
+    strIn.min = "-100"; strIn.max = "100"; strIn.step = "0.01";
+    strIn.value = it.lora_strength; strIn.disabled = !it.enabled; strIn.dataset.k = "it-lora-strength";
+    strIn.oninput = () => patchIdentityTransfer({ lora_strength: parseFloat(strIn.value || "0") });
+    g.append(field("LoRA strength", strIn));
+
+    const gAdv = group(pane, "Advanced");
+    const projSel = el("select"); projSel.dataset.k = "it-projector";
+    projSel.disabled = !it.enabled;
+    const projNone = el("option", null, "None"); projNone.value = "None";
+    if (it.identity_projector === "None") projNone.selected = true;
+    projSel.append(projNone);
+    loraChoices.forEach((c) => {
+      const o = el("option", null, c); o.value = c;
+      if (c === it.identity_projector) o.selected = true;
+      projSel.append(o);
+    });
+    projSel.onchange = () => patchIdentityTransfer({ identity_projector: projSel.value });
+    gAdv.append(field("ArcFace projector", projSel,
+      "Optional secondary identity channel — weak, push id_strength high (5-20) to test."));
+
+    const arcSel = el("select"); arcSel.dataset.k = "it-arcface-mode";
+    arcSel.disabled = !it.enabled;
+    ["auto_adjust", "as_is", "disable"].forEach((c) => {
+      const o = el("option", null, c); o.value = c;
+      if (c === it.arcface_mode) o.selected = true;
+      arcSel.append(o);
+    });
+    arcSel.onchange = () => patchIdentityTransfer({ arcface_mode: arcSel.value });
+    gAdv.append(field("ArcFace detection mode", arcSel));
+
+    const num = (label, val, key, min, max, step) => {
+      const i = el("input"); i.type = "number"; i.min = String(min); i.max = String(max); i.step = String(step);
+      i.value = val; i.disabled = !it.enabled; i.dataset.k = "it-" + key;
+      i.oninput = () => patchIdentityTransfer({ [key]: parseFloat(i.value || "0") });
+      gAdv.append(field(label, i));
+    };
+    num("Source phase id", it.source_id, "source_id", 0, 8, 1);
+    num("Phase scale", it.phase_scale, "phase_scale", 0, 4, 0.1);
+    num("ArcFace token strength", it.id_strength, "id_strength", 0, 50, 0.5);
+
+    const dbgCb = el("input"); dbgCb.type = "checkbox"; dbgCb.checked = it.debug_log;
+    dbgCb.style.width = "auto";
+    dbgCb.disabled = !it.enabled; dbgCb.dataset.k = "it-debug";
+    dbgCb.onchange = () => patchIdentityTransfer({ debug_log: dbgCb.checked });
+    gAdv.append(toggleField("Debug log", dbgCb, "Print per-step [LTXIdOverlap] shape logs to the ComfyUI console."));
+  }
+
   function normGuideSettings(p) {
     const gs = (p && p.guide_settings) || {};
     return { stack_enabled: !!gs.stack_enabled, accumulate_prior: !!gs.accumulate_prior };
@@ -316,6 +466,7 @@
         { id: "chain_timing", group: "Chain Sampler", title: "Timing & Seed", icon: "⏱", badge: countChainView(p, "chain_timing") || null },
         { id: "chain_guidance", group: "Chain Sampler", title: "Guidance", icon: "◇", badge: countChainView(p, "chain_guidance") || null },
         { id: "chain_decode", group: "Chain Sampler", title: "Decode", icon: "▣", badge: countChainView(p, "chain_decode") || null },
+        { id: "chain_identity", group: "Chain Sampler", title: "Identity (Best-FaceID)", icon: "⊙", badge: normIdentityTransfer(p).enabled ? 1 : null },
         { id: "chain_experimental", group: "Chain Sampler", title: "Experimental", icon: "⚗", badge: countChainView(p, "chain_experimental") || null },
       );
     }
@@ -578,6 +729,7 @@
       case "chain_timing": return renderChainTiming(pane, st);
       case "chain_guidance": return renderChainKnobsView(pane, st, "chain_guidance", "Guidance");
       case "chain_decode": return renderChainKnobsView(pane, st, "chain_decode", "Decode");
+      case "chain_identity": return renderChainIdentity(pane, st);
       case "chain_experimental": return renderChainKnobsView(pane, st, "chain_experimental", "Experimental",
         "Research techniques — off by default; expect quality/overhead trade-offs.");
       default: return renderOverview(pane, st);
@@ -642,6 +794,7 @@
     body.append(content);
     _mounted = { content };
     view = "overview";
+    ensureIdentityInfo();
 
     content.addEventListener("focusin", (e) => {
       const t = e.target;
diff --git a/movie_editor/server.py b/movie_editor/server.py
index 13835da0..36705077 100644
--- a/movie_editor/server.py
+++ b/movie_editor/server.py
@@ -378,6 +378,37 @@ def _prepare_media(proj: Project, extra_refs: Optional[list] = None, *, chain_av
     return {"primary": primary, "by_scene": by_scene}
 
 
+def _identity_transfer_params(p: Project) -> Optional[dict]:
+    """Resolve the project's Best-FaceID config into builder.build() params, copying the
+    reference face image into ComfyUI's input dir (same mechanism as scene anchors)."""
+    it = getattr(p, "identity_transfer", None) or {}
+    if not it.get("enabled"):
+        return None
+    ref = it.get("reference_media_ref")
+    if not ref:
+        return None
+    try:
+        import folder_paths
+        indir = folder_paths.get_input_directory()
+    except Exception:
+        return None
+    fn = _copy_scene_media(ref, indir)
+    if not fn:
+        return None
+    return {**it, "reference_filename": fn}
+
+
+def _missing_identity_reference_media(p: Project) -> Optional[str]:
+    """The Best-FaceID reference face asset id, if enabled but missing from the media bin."""
+    it = getattr(p, "identity_transfer", None) or {}
+    if not it.get("enabled"):
+        return None
+    ref = it.get("reference_media_ref")
+    if not ref or not media.path_for(ref):
+        return ref or "(none selected)"
+    return None
+
+
 def _project_or_404(pid: str) -> Project:
     try:
         p = projects.get(pid)
@@ -1589,6 +1620,14 @@ if web is not None and PromptServer is not None:
                 {"detail": f"Missing i2v anchor image in Media bin ({ids}). Re-save the frame or pick another anchor."},
                 status=400,
             )
+        missing_ref = _missing_identity_reference_media(p)
+        if missing_ref:
+            return web.json_response(
+                {"detail": f"Best-FaceID identity transfer is enabled but its reference face "
+                           f"image is missing ({missing_ref}). Pick or re-upload it in "
+                           f"Models → Identity Transfer."},
+                status=400,
+            )
         # V1 uniform chain: if trimmed scenes all agree on a frame count, use it.
         # This makes the timeline trim handle actually affect generation length,
         # provided the user has linked EmptyLTXVLatent.num_frames → Project Frames
@@ -1629,6 +1668,7 @@ if web is not None and PromptServer is not None:
                 "variables": list(target.variables or []),
                 "reset_session": reset_session,
                 "refinement_key": (target.refinement_key or "default"),
+                "identity_transfer": _identity_transfer_params(p),
             }, media=(media_pack or {}).get("primary") if media_pack else None)
         except Exception as e:  # noqa: BLE001
             return web.json_response({"detail": f"Workflow build failed: {e}"}, status=502)
diff --git a/movie_editor/tests/test_builder.py b/movie_editor/tests/test_builder.py
index cc32875e..f7d5c495 100644
--- a/movie_editor/tests/test_builder.py
+++ b/movie_editor/tests/test_builder.py
@@ -353,3 +353,90 @@ def test_bypass_on_node_with_no_matching_input_leaves_it_active_and_reports():
     graph, report = builder.build(OI, models, PARAMS)
     assert "slot_v" in graph
     assert any("bypass needs exactly one input" in u for u in report["unsatisfied"])
+
+
+# ── Best-FaceID identity transfer (ComfyUI-BFSNodes LTX Identity Transfer) ────────────────
+
+IDENTITY_OI = {
+    **OI,
+    "LTXIdentityTransfer": {
+        "input": {"required": {
+            "model": ["MODEL"], "positive": ["CONDITIONING"], "negative": ["CONDITIONING"],
+            "vae": ["VAE"], "latent": ["LATENT"], "reference_face": ["IMAGE"],
+            "identity_projector": [["None", "faceid.safetensors"], {"default": "None"}],
+            "source_id": ["FLOAT", {"default": 2.0}],
+            "phase_scale": ["FLOAT", {"default": 1.0}],
+            "id_strength": ["FLOAT", {"default": 1.0}],
+            "arcface_mode": [["auto_adjust", "as_is", "disable"], {"default": "auto_adjust"}],
+            "debug_log": ["BOOLEAN", {"default": False}],
+        }},
+        "output": ["MODEL", "CONDITIONING", "CONDITIONING", "LATENT", "STRING"],
+        "output_name": ["model", "positive", "negative", "latent", "debug"],
+    },
+    "LoraLoaderModelOnly": {
+        "input": {"required": {"model": ["MODEL"], "lora_name": [["faceid.safetensors"]],
+                                "strength_model": ["FLOAT", {"default": 1.0}]}},
+        "output": ["MODEL"], "output_name": ["MODEL"],
+    },
+}
+
+IDENTITY_MODELS = {"full_control": True, "slots": [
+    {"id": "u", "node_class": "UnetLoader", "inputs": {"unet_name": "m.safetensors"},
+     "wires": {"MODEL": "port:FunPackStudio.model"}},
+    {"id": "c", "node_class": "ClipLoader", "inputs": {}, "wires": {"CLIP": "port:FunPackStudio.clip"}},
+    {"id": "v", "node_class": "VaeLoader", "inputs": {}, "wires": {}},
+]}
+
+
+def test_identity_transfer_splices_between_studio_and_sampler():
+    params = {**PARAMS, "identity_transfer": {
+        "enabled": True, "lora_name": "faceid.safetensors", "reference_filename": "ref.png"}}
+    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, params)
+    assert graph["id_ref_load"]["inputs"]["image"] == "ref.png"
+    assert graph["id_lora"]["inputs"]["model"] == ["slot_u", 0]
+    assert graph["id_lora"]["inputs"]["lora_name"] == "faceid.safetensors"
+    assert graph["studio"]["inputs"]["model"] == ["id_lora", 0]
+    assert graph["idxfer"]["inputs"]["model"] == ["studio", 0]
+    assert graph["idxfer"]["inputs"]["positive"] == ["cond", 0]
+    assert graph["idxfer"]["inputs"]["negative"] == ["cond", 1]
+    assert graph["idxfer"]["inputs"]["latent"] == ["studio", 12]
+    assert graph["idxfer"]["inputs"]["reference_face"] == ["id_ref_load", 0]
+    assert graph["idxfer"]["inputs"]["vae"] == ["slot_v", 0]
+    assert graph["sampler"]["inputs"]["model"] == ["idxfer", 0]
+    assert graph["sampler"]["inputs"]["positive"] == ["idxfer", 1]
+    assert graph["sampler"]["inputs"]["negative"] == ["idxfer", 2]
+    assert graph["concat"]["inputs"]["video_latent"] == ["idxfer", 3]
+    assert not any("Identity Transfer" in b for b in report["blocking"])
+
+
+def test_identity_transfer_without_lora_name_skips_lora_insertion():
+    # The LoRA is a separate convenience — Identity Transfer itself still works if the user
+    # already loaded the LoRA another way (or is testing without it).
+    params = {**PARAMS, "identity_transfer": {"enabled": True, "reference_filename": "ref.png"}}
+    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, params)
+    assert "id_lora" not in graph
+    assert graph["studio"]["inputs"]["model"] == ["slot_u", 0]  # untouched
+    assert graph["idxfer"]["inputs"]["model"] == ["studio", 0]
+    assert not any("Identity Transfer" in b for b in report["blocking"])
+
+
+def test_identity_transfer_missing_node_class_blocks():
+    params = {**PARAMS, "identity_transfer": {
+        "enabled": True, "lora_name": "faceid.safetensors", "reference_filename": "ref.png"}}
+    graph, report = builder.build(OI, IDENTITY_MODELS, params)  # OI has no identity node installed
+    assert "idxfer" not in graph
+    assert any("LTX Identity Transfer" in b for b in report["blocking"])
+
+
+def test_identity_transfer_missing_reference_blocks():
+    params = {**PARAMS, "identity_transfer": {"enabled": True, "lora_name": "faceid.safetensors"}}
+    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, params)
+    assert "idxfer" not in graph
+    assert any("reference face" in b for b in report["blocking"])
+
+
+def test_identity_transfer_disabled_by_default():
+    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, PARAMS)
+    assert "idxfer" not in graph
+    assert "id_lora" not in graph
+    assert not any("Identity Transfer" in b for b in report["blocking"])

From 622e8ef747c17a0b60c99f6fb1d485228f8ad122 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 20:46:56 +0300
Subject: [PATCH 09/11] Best-FaceID identity transfer: native source-phase RoPE
 port, no external node

Reverts the earlier ComfyUI-BFSNodes graph-splice approach (builder.py/timeline.py/
server.py scaffolding) in favor of porting the missing piece natively: FunPack's own
guide-attention path (LTXVAddGuide.add_keyframe_index + guide_attention_entries)
already places a reference image on the frame-0 grid exactly like BFSNodes' overlap
technique, and Continuity's existing "Identity pin" already applies one reference
face to every scene. The only real gap was the LoRA's trained-time source_phase RoPE
tag, added as a tag+strip patch on FunPackLTXAVSceneChainSampler scoped to the exact
guide-entry token range (never co-active mid_scene/carry/prior-scene guides sharing
the same keyframe_idxs tensor). New "Best-FaceID compatibility" + "Source-phase id"
knobs live under Engine Settings -> Experimental, alongside every other unvalidated
technique. LoRA loading itself reuses the existing generic Models-menu lora role.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---
 movie_editor/backend/builder.py          |  83 --------
 movie_editor/backend/timeline.py         |  20 +-
 movie_editor/frontend/engine_settings.js | 161 +--------------
 movie_editor/server.py                   |  40 ----
 movie_editor/tests/test_builder.py       |  87 ---------
 samplers.py                              | 199 ++++++++++++++++++-
 tests/test_identity_transfer.py          | 239 +++++++++++++++++++++++
 7 files changed, 445 insertions(+), 384 deletions(-)
 create mode 100644 tests/test_identity_transfer.py

diff --git a/movie_editor/backend/builder.py b/movie_editor/backend/builder.py
index 101f728f..3e80f6f1 100644
--- a/movie_editor/backend/builder.py
+++ b/movie_editor/backend/builder.py
@@ -93,20 +93,6 @@ CORE_PRODUCERS: list[tuple[str, int, str]] = [  # (core_id, output_index, type)
     ("f2i", 0, "INT"),
 ]
 
-# Best-FaceID identity transfer (ComfyUI-BFSNodes): a single project-wide splice, not a
-# generic Models-menu slot — it needs simultaneous model+positive+negative+vae+latent+
-# reference_face, a shape no ROLE covers. Detected by class presence so the feature
-# disappears cleanly when the custom node package isn't installed. Both names alias the
-# same node class (the package kept the old key as a back-compat alias).
-IDENTITY_TRANSFER_CLASSES: tuple[str, ...] = ("LTXIdentityTransfer", "LTXIdentityOverlapConditioning")
-
-
-def identity_transfer_class(object_info: dict) -> Optional[str]:
-    for cls in IDENTITY_TRANSFER_CLASSES:
-        if cls in (object_info or {}):
-            return cls
-    return None
-
 CONTROL_VALUES = {"fixed", "randomize", "increment", "decrement"}
 # Type-appropriate empty values for widgets that declare no default.
 _WIDGET_EMPTY = {"STRING": "", "INT": 0, "FLOAT": 0.0, "BOOLEAN": False}
@@ -556,21 +542,6 @@ def build(object_info: dict, models_config: dict, params: dict, media: dict | No
     # dropped here; an all-explicit loop is reported precisely. Never hand ComfyUI a cyclic graph.
     _break_cycles(graph, report, protected_edges, _node_labels(slots, slot_node_id, object_info))
 
-    # 6. Best-FaceID identity transfer: a fixed-core splice (not a slot), applied last so it
-    # can tap the already-resolved Studio · model and sampler · vae sources.
-    it_cfg = params.get("identity_transfer") or {}
-    if not disable_core and it_cfg.get("enabled"):
-        it_cls = identity_transfer_class(object_info)
-        if not it_cls:
-            msg = ("Identity Transfer (Best-FaceID) is enabled but no LTX Identity Transfer "
-                   "node is installed — install ComfyUI-BFSNodes.")
-            report["unsatisfied"].append(msg); report["blocking"].append(msg)
-        elif not it_cfg.get("reference_filename"):
-            msg = "Identity Transfer (Best-FaceID) is enabled but no reference face image is set."
-            report["unsatisfied"].append(msg); report["blocking"].append(msg)
-        else:
-            _splice_identity_transfer(graph, object_info, it_cfg, it_cls, report)
-
     for msg in pipeline_wiring.validate_models_wiring(models_config):
         report["blocking"].append(msg)
         report["unsatisfied"].append(msg)
@@ -935,57 +906,3 @@ def _apply_bypass(graph, slots, slot_node_id, slot_def, report):
                         del ndata["inputs"][inp_name]
         del graph[sid]
         report["wired"].append(f"{s.get('node_class')} bypassed (pass-through)")
-
-
-def _splice_identity_transfer(
-    graph: dict, object_info: dict, cfg: dict, cls: str, report: dict[str, list],
-) -> None:
-    """Insert the Best-FaceID LoRA + LTX Identity Transfer node between the fixed core's
-    loader->Studio edge and Studio/cond->sampler/concat edges.
-
-    Runs after every other wiring step, so it taps the already-resolved Studio · model and
-    sampler · vae sources and always wins over whatever fed sampler.model/positive/negative
-    and concat.video_latent — Identity Transfer isn't meant to be combined with manually
-    rewiring those exact ports in Full control.
-    """
-    lora_name = cfg.get("lora_name")
-    model_src = graph["studio"]["inputs"].get("model")
-    if lora_name:
-        lora_cls = "LoraLoaderModelOnly"
-        if lora_cls not in object_info:
-            msg = (f"Identity Transfer needs '{lora_cls}' (built into ComfyUI) but it's "
-                   "missing from the node registry.")
-            report["unsatisfied"].append(msg); report["blocking"].append(msg)
-        else:
-            lora_inputs = _widget_defaults(object_info.get(lora_cls))
-            lora_inputs["model"] = model_src
-            lora_inputs["lora_name"] = lora_name
-            lora_inputs["strength_model"] = float(cfg.get("lora_strength", 1.0))
-            graph["id_lora"] = {"class_type": lora_cls, "inputs": lora_inputs}
-            graph["studio"]["inputs"]["model"] = ["id_lora", 0]
-            report["wired"].append(f"Best-FaceID LoRA ({lora_name}) -> Studio · model")
-
-    graph["id_ref_load"] = {
-        "class_type": "LoadImage",
-        "inputs": {"image": cfg["reference_filename"]},
-    }
-
-    inputs = _widget_defaults(object_info.get(cls))
-    inputs.update({
-        "model": ["studio", 0],
-        "positive": ["cond", 0],
-        "negative": ["cond", 1],
-        "vae": graph["sampler"]["inputs"].get("vae"),
-        "latent": ["studio", 12],
-        "reference_face": ["id_ref_load", 0],
-    })
-    for k in ("identity_projector", "source_id", "phase_scale", "id_strength", "arcface_mode", "debug_log"):
-        if cfg.get(k) is not None:
-            inputs[k] = cfg[k]
-    graph["idxfer"] = {"class_type": cls, "inputs": inputs}
-
-    graph["sampler"]["inputs"]["model"] = ["idxfer", 0]
-    graph["sampler"]["inputs"]["positive"] = ["idxfer", 1]
-    graph["sampler"]["inputs"]["negative"] = ["idxfer", 2]
-    graph["concat"]["inputs"]["video_latent"] = ["idxfer", 3]
-    report["wired"].append("LTX Identity Transfer spliced between Studio and the sampler")
diff --git a/movie_editor/backend/timeline.py b/movie_editor/backend/timeline.py
index a55d859b..43d17252 100644
--- a/movie_editor/backend/timeline.py
+++ b/movie_editor/backend/timeline.py
@@ -101,8 +101,8 @@ def _anchor_media_ref(scene: Scene) -> Optional[str]:
     return None
 
 
-def _identity_pin_guide(media_ref: str, strength: float) -> dict:
-    return {
+def _identity_pin_guide(media_ref: str, strength: float, *, is_pin: bool = True) -> dict:
+    entry = {
         "enabled": True,
         "source": "image",
         "media_ref": media_ref,
@@ -110,12 +110,19 @@ def _identity_pin_guide(media_ref: str, strength: float) -> dict:
         "apply_at": 0,
         "strength": max(0.25, min(0.5, float(strength))),
     }
+    if is_pin:
+        # Marks this entry (not prior-scene/mid-scene/template guides) as the one eligible
+        # for the sampler's identity_transfer source-phase RoPE tag — see samplers.py.
+        entry["identity_pin"] = True
+    return entry
 
 
 def _prior_scene_guide(prior: Scene, strength: float) -> Optional[dict]:
     ref = _anchor_media_ref(prior)
     if ref:
-        return _identity_pin_guide(ref, strength)
+        # Reuses the same image-guide shape as the project identity pin, but this is the
+        # PRIOR SCENE's own anchor (motion/layout continuity) — never the identity_pin tag.
+        return _identity_pin_guide(ref, strength, is_pin=False)
     return {
         "enabled": True,
         "source": "template",
@@ -595,12 +602,6 @@ class Project:
     # (same shape as the global models.json). Empty {"slots": []} falls back to the
     # global default at build/read time; the editor seeds new projects from it.
     models: dict = field(default_factory=lambda: {"slots": []})
-    # Best-FaceID identity transfer (ComfyUI-BFSNodes' LTX Identity Transfer node), a single
-    # project-wide splice: {enabled, reference_media_ref, lora_name, lora_strength,
-    # identity_projector, source_id, phase_scale, id_strength, arcface_mode, debug_log}.
-    # One reference face for the whole project — the chain sampler runs multi-scene prompts
-    # through one graph per run, so there's no per-scene seam to hang a second identity on.
-    identity_transfer: dict = field(default_factory=dict)
     # Guide stack toggles — empty / stack_enabled=false keeps Studio carry behaviour.
     guide_settings: dict = field(default_factory=dict)
     continuity_settings: dict = field(default_factory=dict)
@@ -650,7 +651,6 @@ class Project:
             overlay_lanes=list(d.get("overlay_lanes") or []),
             overlay_tracks=list(d.get("overlay_tracks") or []),
             models=dict(d.get("models") or {"slots": []}),
-            identity_transfer=dict(d.get("identity_transfer") or {}),
             guide_settings=dict(d.get("guide_settings") or {}),
             continuity_settings=dict(d.get("continuity_settings") or {}),
             generation_meta=dict(d.get("generation_meta") or {}),
diff --git a/movie_editor/frontend/engine_settings.js b/movie_editor/frontend/engine_settings.js
index 2ff2e221..6eea3300 100644
--- a/movie_editor/frontend/engine_settings.js
+++ b/movie_editor/frontend/engine_settings.js
@@ -5,7 +5,6 @@
 (function () {
   const { el, clear } = window.dom;
   const S = window.Store;
-  const API = window.MovieEditorAPI;
 
   let _mounted = null; // { scroller (pane, set per render), content (shell root) }
   let unsub = null;
@@ -180,6 +179,10 @@
     { name: "dynashift",             label: "DynaShift (steer off bad gens)", kind: "bool", default: false },
     { name: "dynashift_strength",    label: "DynaShift strength",    kind: "float", default: 0.3, min: 0.05, max: 1.0, step: 0.05, dependsOn: "dynashift" },
     { name: "dynashift_threshold",   label: "DynaShift match threshold", kind: "float", default: 0.6, min: 0.3, max: 0.95, step: 0.05, dependsOn: "dynashift" },
+    { name: "identity_transfer_enabled", label: "Best-FaceID compatibility", kind: "bool", default: false,
+      hint: "Tags Continuity's Identity pin guide (Engine → Continuity) with the source-phase RoPE rotation Best-FaceID-style identity LoRAs were trained on. Load the LoRA itself the normal way — Models → add a LoRA loader onto the model path. No effect without an Identity pin image set." },
+    { name: "identity_transfer_phase", label: "Source-phase id", kind: "float", default: 2.0, min: 0.0, max: 8.0, step: 0.5, dependsOn: "identity_transfer_enabled",
+      hint: "Matches the LoRA's training convention (ltx-trainer used 2)." },
   ];
   const SAMPLER_KNOB_MAP = Object.fromEntries(SAMPLER_KNOBS.map((k) => [k.name, k]));
 
@@ -215,155 +218,6 @@
     S.patchProject({ continuity_settings: { ...(p.continuity_settings || {}), ...patch } });
   }
 
-  // ── Best-FaceID identity transfer (ComfyUI-BFSNodes LTX Identity Transfer) ────────────
-  const IDENTITY_TRANSFER_CLASSES = ["LTXIdentityTransfer", "LTXIdentityOverlapConditioning"];
-  let idNodesInfo = null; // { available, loraChoices } — fetched lazily, cached for the session
-
-  async function ensureIdentityInfo(force) {
-    if (idNodesInfo && !force) return idNodesInfo;
-    const info = { available: false, loraChoices: [] };
-    try {
-      const nodes = (await API.allNodes()).nodes || [];
-      info.available = nodes.some((n) => IDENTITY_TRANSFER_CLASSES.includes(n.class));
-      if (info.available) {
-        const lora = await API.nodeSpec("LoraLoaderModelOnly");
-        const w = (lora?.inputs || []).find((i) => i.name === "lora_name");
-        info.loraChoices = (w && w.choices) || [];
-      }
-    } catch (_) { /* leave unavailable */ }
-    idNodesInfo = info;
-    render();
-    return info;
-  }
-
-  function normIdentityTransfer(p) {
-    const it = (p && p.identity_transfer) || {};
-    return {
-      enabled: !!it.enabled,
-      reference_media_ref: it.reference_media_ref || null,
-      lora_name: it.lora_name || "",
-      lora_strength: it.lora_strength != null ? +it.lora_strength : 1.0,
-      identity_projector: it.identity_projector || "None",
-      source_id: it.source_id != null ? +it.source_id : 2.0,
-      phase_scale: it.phase_scale != null ? +it.phase_scale : 1.0,
-      id_strength: it.id_strength != null ? +it.id_strength : 1.0,
-      arcface_mode: it.arcface_mode || "auto_adjust",
-      debug_log: !!it.debug_log,
-    };
-  }
-
-  function patchIdentityTransfer(patch) {
-    const p = S.get().project;
-    S.patchProject({ identity_transfer: { ...(p.identity_transfer || {}), ...patch } });
-  }
-
-  function renderChainIdentity(pane, st) {
-    const p = st.project;
-    const it = normIdentityTransfer(p);
-    const info = idNodesInfo;
-    if (!info) {
-      pane.append(hintEl("Checking for ComfyUI-BFSNodes…"));
-      return;
-    }
-    if (!info.available) {
-      pane.append(hintEl(
-        "Best-FaceID identity transfer needs the LTX Identity Transfer node from "
-        + "ComfyUI-BFSNodes (github.com/alisson-anjos/ComfyUI-BFSNodes) — install it in "
-        + "custom_nodes and restart ComfyUI."));
-      const retry = el("button", "btn ghost tiny", "Check again");
-      retry.type = "button";
-      retry.onclick = () => ensureIdentityInfo(true);
-      pane.append(retry);
-      return;
-    }
-    pane.append(hintEl(
-      "One reference face for the whole project. Loads the Best-FaceID LoRA on the model "
-      + "path and feeds the reference image into LTX Identity Transfer, spliced between "
-      + "Studio and the Chain Sampler. Tip: prefix your prompt with “ref_t2v:” for "
-      + "best results."));
-
-    const g = group(pane, "Best-FaceID");
-    const enCb = el("input"); enCb.type = "checkbox"; enCb.checked = it.enabled;
-    enCb.style.width = "auto";
-    enCb.dataset.k = "it-enabled";
-    enCb.onchange = () => patchIdentityTransfer({ enabled: enCb.checked });
-    g.append(toggleField("Enable identity transfer", enCb));
-
-    const refRow = el("div", "sw-row eng-field eng-stack");
-    const refMain = el("div", "sw-row-main");
-    refMain.append(el("div", "sw-row-title", "Reference face"));
-    refRow.append(refMain);
-    const refPicker = window.MediaPicker.create({
-      value: it.reference_media_ref,
-      mediaBin: st.mediaBin,
-      noneLabel: "— pick a reference face —",
-      onChange: (v) => patchIdentityTransfer({ reference_media_ref: v }),
-    });
-    if (!it.enabled) refPicker.classList.add("disabled");
-    refRow.append(refPicker);
-    g.append(refRow);
-
-    const loraChoices = info.loraChoices || [];
-    const loraSel = el("select"); loraSel.dataset.k = "it-lora";
-    loraSel.disabled = !it.enabled;
-    const noneOpt = el("option", null, "— select LoRA —"); noneOpt.value = "";
-    loraSel.append(noneOpt);
-    loraChoices.forEach((c) => {
-      const o = el("option", null, c); o.value = c;
-      if (c === it.lora_name) o.selected = true;
-      loraSel.append(o);
-    });
-    loraSel.onchange = () => patchIdentityTransfer({ lora_name: loraSel.value });
-    g.append(field("Best-FaceID LoRA", loraSel));
-
-    const strIn = el("input"); strIn.type = "number";
-    strIn.min = "-100"; strIn.max = "100"; strIn.step = "0.01";
-    strIn.value = it.lora_strength; strIn.disabled = !it.enabled; strIn.dataset.k = "it-lora-strength";
-    strIn.oninput = () => patchIdentityTransfer({ lora_strength: parseFloat(strIn.value || "0") });
-    g.append(field("LoRA strength", strIn));
-
-    const gAdv = group(pane, "Advanced");
-    const projSel = el("select"); projSel.dataset.k = "it-projector";
-    projSel.disabled = !it.enabled;
-    const projNone = el("option", null, "None"); projNone.value = "None";
-    if (it.identity_projector === "None") projNone.selected = true;
-    projSel.append(projNone);
-    loraChoices.forEach((c) => {
-      const o = el("option", null, c); o.value = c;
-      if (c === it.identity_projector) o.selected = true;
-      projSel.append(o);
-    });
-    projSel.onchange = () => patchIdentityTransfer({ identity_projector: projSel.value });
-    gAdv.append(field("ArcFace projector", projSel,
-      "Optional secondary identity channel — weak, push id_strength high (5-20) to test."));
-
-    const arcSel = el("select"); arcSel.dataset.k = "it-arcface-mode";
-    arcSel.disabled = !it.enabled;
-    ["auto_adjust", "as_is", "disable"].forEach((c) => {
-      const o = el("option", null, c); o.value = c;
-      if (c === it.arcface_mode) o.selected = true;
-      arcSel.append(o);
-    });
-    arcSel.onchange = () => patchIdentityTransfer({ arcface_mode: arcSel.value });
-    gAdv.append(field("ArcFace detection mode", arcSel));
-
-    const num = (label, val, key, min, max, step) => {
-      const i = el("input"); i.type = "number"; i.min = String(min); i.max = String(max); i.step = String(step);
-      i.value = val; i.disabled = !it.enabled; i.dataset.k = "it-" + key;
-      i.oninput = () => patchIdentityTransfer({ [key]: parseFloat(i.value || "0") });
-      gAdv.append(field(label, i));
-    };
-    num("Source phase id", it.source_id, "source_id", 0, 8, 1);
-    num("Phase scale", it.phase_scale, "phase_scale", 0, 4, 0.1);
-    num("ArcFace token strength", it.id_strength, "id_strength", 0, 50, 0.5);
-
-    const dbgCb = el("input"); dbgCb.type = "checkbox"; dbgCb.checked = it.debug_log;
-    dbgCb.style.width = "auto";
-    dbgCb.disabled = !it.enabled; dbgCb.dataset.k = "it-debug";
-    dbgCb.onchange = () => patchIdentityTransfer({ debug_log: dbgCb.checked });
-    gAdv.append(toggleField("Debug log", dbgCb, "Print per-step [LTXIdOverlap] shape logs to the ComfyUI console."));
-  }
-
   function normGuideSettings(p) {
     const gs = (p && p.guide_settings) || {};
     return { stack_enabled: !!gs.stack_enabled, accumulate_prior: !!gs.accumulate_prior };
@@ -415,7 +269,7 @@
         S.setSamplerInput(k.name, v);
       };
     }
-    parentGroup.append(field(k.label + (forced ? " (auto)" : ""), ctrl));
+    parentGroup.append(field(k.label + (forced ? " (auto)" : ""), ctrl, k.hint));
   }
 
   function renderKnobList(parentGroup, st, names) {
@@ -433,7 +287,7 @@
     chain_timing: ["frame_overlap", "transition_duration", "use_same_seed"],
     chain_guidance: ["cfg", "embed_guidance", "embed_guidance_source", "embed_guidance_strength", "score_slider", "score_slider_strength", "output_guidance", "output_guidance_strength", "dynashift", "dynashift_strength", "dynashift_threshold"],
     chain_decode: ["decode_noise_scale", "decode_timestep", "decode_tile_size"],
-    chain_experimental: ["mid_scene_guide", "mid_scene_guide_strength", "joyai_memory", "joyai_memory_size", "joyai_fix_frames", "joyai_frame_select", "joyai_memory_strength", "joyai_audio_memory", "v2a_grad_scale", "alg_blur_guides", "alg_guide_blur_strength", "alg_guide_blur_sigma_threshold", "bounded_attention_enabled"],
+    chain_experimental: ["mid_scene_guide", "mid_scene_guide_strength", "joyai_memory", "joyai_memory_size", "joyai_fix_frames", "joyai_frame_select", "joyai_memory_strength", "joyai_audio_memory", "v2a_grad_scale", "alg_blur_guides", "alg_guide_blur_strength", "alg_guide_blur_sigma_threshold", "bounded_attention_enabled", "identity_transfer_enabled", "identity_transfer_phase"],
   };
 
   function countChainView(p, id) {
@@ -466,7 +320,6 @@
         { id: "chain_timing", group: "Chain Sampler", title: "Timing & Seed", icon: "⏱", badge: countChainView(p, "chain_timing") || null },
         { id: "chain_guidance", group: "Chain Sampler", title: "Guidance", icon: "◇", badge: countChainView(p, "chain_guidance") || null },
         { id: "chain_decode", group: "Chain Sampler", title: "Decode", icon: "▣", badge: countChainView(p, "chain_decode") || null },
-        { id: "chain_identity", group: "Chain Sampler", title: "Identity (Best-FaceID)", icon: "⊙", badge: normIdentityTransfer(p).enabled ? 1 : null },
         { id: "chain_experimental", group: "Chain Sampler", title: "Experimental", icon: "⚗", badge: countChainView(p, "chain_experimental") || null },
       );
     }
@@ -729,7 +582,6 @@
       case "chain_timing": return renderChainTiming(pane, st);
       case "chain_guidance": return renderChainKnobsView(pane, st, "chain_guidance", "Guidance");
       case "chain_decode": return renderChainKnobsView(pane, st, "chain_decode", "Decode");
-      case "chain_identity": return renderChainIdentity(pane, st);
       case "chain_experimental": return renderChainKnobsView(pane, st, "chain_experimental", "Experimental",
         "Research techniques — off by default; expect quality/overhead trade-offs.");
       default: return renderOverview(pane, st);
@@ -794,7 +646,6 @@
     body.append(content);
     _mounted = { content };
     view = "overview";
-    ensureIdentityInfo();
 
     content.addEventListener("focusin", (e) => {
       const t = e.target;
diff --git a/movie_editor/server.py b/movie_editor/server.py
index 36705077..13835da0 100644
--- a/movie_editor/server.py
+++ b/movie_editor/server.py
@@ -378,37 +378,6 @@ def _prepare_media(proj: Project, extra_refs: Optional[list] = None, *, chain_av
     return {"primary": primary, "by_scene": by_scene}
 
 
-def _identity_transfer_params(p: Project) -> Optional[dict]:
-    """Resolve the project's Best-FaceID config into builder.build() params, copying the
-    reference face image into ComfyUI's input dir (same mechanism as scene anchors)."""
-    it = getattr(p, "identity_transfer", None) or {}
-    if not it.get("enabled"):
-        return None
-    ref = it.get("reference_media_ref")
-    if not ref:
-        return None
-    try:
-        import folder_paths
-        indir = folder_paths.get_input_directory()
-    except Exception:
-        return None
-    fn = _copy_scene_media(ref, indir)
-    if not fn:
-        return None
-    return {**it, "reference_filename": fn}
-
-
-def _missing_identity_reference_media(p: Project) -> Optional[str]:
-    """The Best-FaceID reference face asset id, if enabled but missing from the media bin."""
-    it = getattr(p, "identity_transfer", None) or {}
-    if not it.get("enabled"):
-        return None
-    ref = it.get("reference_media_ref")
-    if not ref or not media.path_for(ref):
-        return ref or "(none selected)"
-    return None
-
-
 def _project_or_404(pid: str) -> Project:
     try:
         p = projects.get(pid)
@@ -1620,14 +1589,6 @@ if web is not None and PromptServer is not None:
                 {"detail": f"Missing i2v anchor image in Media bin ({ids}). Re-save the frame or pick another anchor."},
                 status=400,
             )
-        missing_ref = _missing_identity_reference_media(p)
-        if missing_ref:
-            return web.json_response(
-                {"detail": f"Best-FaceID identity transfer is enabled but its reference face "
-                           f"image is missing ({missing_ref}). Pick or re-upload it in "
-                           f"Models → Identity Transfer."},
-                status=400,
-            )
         # V1 uniform chain: if trimmed scenes all agree on a frame count, use it.
         # This makes the timeline trim handle actually affect generation length,
         # provided the user has linked EmptyLTXVLatent.num_frames → Project Frames
@@ -1668,7 +1629,6 @@ if web is not None and PromptServer is not None:
                 "variables": list(target.variables or []),
                 "reset_session": reset_session,
                 "refinement_key": (target.refinement_key or "default"),
-                "identity_transfer": _identity_transfer_params(p),
             }, media=(media_pack or {}).get("primary") if media_pack else None)
         except Exception as e:  # noqa: BLE001
             return web.json_response({"detail": f"Workflow build failed: {e}"}, status=502)
diff --git a/movie_editor/tests/test_builder.py b/movie_editor/tests/test_builder.py
index f7d5c495..cc32875e 100644
--- a/movie_editor/tests/test_builder.py
+++ b/movie_editor/tests/test_builder.py
@@ -353,90 +353,3 @@ def test_bypass_on_node_with_no_matching_input_leaves_it_active_and_reports():
     graph, report = builder.build(OI, models, PARAMS)
     assert "slot_v" in graph
     assert any("bypass needs exactly one input" in u for u in report["unsatisfied"])
-
-
-# ── Best-FaceID identity transfer (ComfyUI-BFSNodes LTX Identity Transfer) ────────────────
-
-IDENTITY_OI = {
-    **OI,
-    "LTXIdentityTransfer": {
-        "input": {"required": {
-            "model": ["MODEL"], "positive": ["CONDITIONING"], "negative": ["CONDITIONING"],
-            "vae": ["VAE"], "latent": ["LATENT"], "reference_face": ["IMAGE"],
-            "identity_projector": [["None", "faceid.safetensors"], {"default": "None"}],
-            "source_id": ["FLOAT", {"default": 2.0}],
-            "phase_scale": ["FLOAT", {"default": 1.0}],
-            "id_strength": ["FLOAT", {"default": 1.0}],
-            "arcface_mode": [["auto_adjust", "as_is", "disable"], {"default": "auto_adjust"}],
-            "debug_log": ["BOOLEAN", {"default": False}],
-        }},
-        "output": ["MODEL", "CONDITIONING", "CONDITIONING", "LATENT", "STRING"],
-        "output_name": ["model", "positive", "negative", "latent", "debug"],
-    },
-    "LoraLoaderModelOnly": {
-        "input": {"required": {"model": ["MODEL"], "lora_name": [["faceid.safetensors"]],
-                                "strength_model": ["FLOAT", {"default": 1.0}]}},
-        "output": ["MODEL"], "output_name": ["MODEL"],
-    },
-}
-
-IDENTITY_MODELS = {"full_control": True, "slots": [
-    {"id": "u", "node_class": "UnetLoader", "inputs": {"unet_name": "m.safetensors"},
-     "wires": {"MODEL": "port:FunPackStudio.model"}},
-    {"id": "c", "node_class": "ClipLoader", "inputs": {}, "wires": {"CLIP": "port:FunPackStudio.clip"}},
-    {"id": "v", "node_class": "VaeLoader", "inputs": {}, "wires": {}},
-]}
-
-
-def test_identity_transfer_splices_between_studio_and_sampler():
-    params = {**PARAMS, "identity_transfer": {
-        "enabled": True, "lora_name": "faceid.safetensors", "reference_filename": "ref.png"}}
-    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, params)
-    assert graph["id_ref_load"]["inputs"]["image"] == "ref.png"
-    assert graph["id_lora"]["inputs"]["model"] == ["slot_u", 0]
-    assert graph["id_lora"]["inputs"]["lora_name"] == "faceid.safetensors"
-    assert graph["studio"]["inputs"]["model"] == ["id_lora", 0]
-    assert graph["idxfer"]["inputs"]["model"] == ["studio", 0]
-    assert graph["idxfer"]["inputs"]["positive"] == ["cond", 0]
-    assert graph["idxfer"]["inputs"]["negative"] == ["cond", 1]
-    assert graph["idxfer"]["inputs"]["latent"] == ["studio", 12]
-    assert graph["idxfer"]["inputs"]["reference_face"] == ["id_ref_load", 0]
-    assert graph["idxfer"]["inputs"]["vae"] == ["slot_v", 0]
-    assert graph["sampler"]["inputs"]["model"] == ["idxfer", 0]
-    assert graph["sampler"]["inputs"]["positive"] == ["idxfer", 1]
-    assert graph["sampler"]["inputs"]["negative"] == ["idxfer", 2]
-    assert graph["concat"]["inputs"]["video_latent"] == ["idxfer", 3]
-    assert not any("Identity Transfer" in b for b in report["blocking"])
-
-
-def test_identity_transfer_without_lora_name_skips_lora_insertion():
-    # The LoRA is a separate convenience — Identity Transfer itself still works if the user
-    # already loaded the LoRA another way (or is testing without it).
-    params = {**PARAMS, "identity_transfer": {"enabled": True, "reference_filename": "ref.png"}}
-    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, params)
-    assert "id_lora" not in graph
-    assert graph["studio"]["inputs"]["model"] == ["slot_u", 0]  # untouched
-    assert graph["idxfer"]["inputs"]["model"] == ["studio", 0]
-    assert not any("Identity Transfer" in b for b in report["blocking"])
-
-
-def test_identity_transfer_missing_node_class_blocks():
-    params = {**PARAMS, "identity_transfer": {
-        "enabled": True, "lora_name": "faceid.safetensors", "reference_filename": "ref.png"}}
-    graph, report = builder.build(OI, IDENTITY_MODELS, params)  # OI has no identity node installed
-    assert "idxfer" not in graph
-    assert any("LTX Identity Transfer" in b for b in report["blocking"])
-
-
-def test_identity_transfer_missing_reference_blocks():
-    params = {**PARAMS, "identity_transfer": {"enabled": True, "lora_name": "faceid.safetensors"}}
-    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, params)
-    assert "idxfer" not in graph
-    assert any("reference face" in b for b in report["blocking"])
-
-
-def test_identity_transfer_disabled_by_default():
-    graph, report = builder.build(IDENTITY_OI, IDENTITY_MODELS, PARAMS)
-    assert "idxfer" not in graph
-    assert "id_lora" not in graph
-    assert not any("Identity Transfer" in b for b in report["blocking"])
diff --git a/samplers.py b/samplers.py
index c91e304c..6b79d5db 100644
--- a/samplers.py
+++ b/samplers.py
@@ -3,6 +3,7 @@ import hashlib
 import math
 import os
 import time as _time
+import types
 
 import torch
 
@@ -2395,6 +2396,14 @@ class FunPackLTXAVSceneChainSampler:
                     "default": 0.975, "min": 0.5, "max": 0.999, "step": 0.005,
                     "tooltip": "Guide frames stay blurred while sigma is above this value (the near-pure-noise steps), then swap to sharp. Higher = narrower blurred window. Independent of the sampler's anchor alg_sigma_threshold.",
                 }),
+                "identity_transfer_enabled": ("BOOLEAN", {
+                    "default": False,
+                    "tooltip": "EXPERIMENTAL Best-FaceID compatibility: tags Continuity's 'Identity pin' guide (Engine settings) with an extra source-phase RoPE rotation, matching the exact overlap+source_phase conditioning Best-FaceID-style identity LoRAs were trained on. No-op without an identity pin image set. Load the LoRA itself the normal way (Models -> add a LoRA loader onto the model path).",
+                }),
+                "identity_transfer_phase": ("FLOAT", {
+                    "default": 2.0, "min": 0.0, "max": 8.0, "step": 0.5,
+                    "tooltip": "Source-phase segment id for the identity-pin guide tokens (ltx-trainer's overlap+source_phase convention used 2). 0 disables the rotation while leaving the guide itself active.",
+                }),
             },
             "hidden": {
                 "unique_id": "UNIQUE_ID",
@@ -3567,8 +3576,34 @@ class FunPackLTXAVSceneChainSampler:
             return 0, True
         return 1 + (at - 1) * time_scale, False
 
-    def _append_guide_latent(self, chunk, guide_frame, apply_at, strength, positive, negative, vae):
-        """Append one guide latent frame with LTX guide attention at apply_at."""
+    def _tag_last_guide_entry(self, cond, key, value):
+        """Best-effort: stamp `key=value` onto the guide_attention_entry _append_guide_attention_entry
+        just added (the last one), without touching any earlier entry. Used to mark which guide
+        (e.g. the Best-FaceID identity pin) is eligible for the source-phase RoPE patch — see
+        _install_identity_phase. No-op (returns cond unchanged) if node_helpers isn't importable
+        or no entries are present; never blocks generation."""
+        try:
+            import node_helpers
+        except ImportError:
+            return cond
+        try:
+            for t in cond:
+                entries = t[1].get("guide_attention_entries")
+                if entries:
+                    tagged = [*entries[:-1], {**entries[-1], key: value}]
+                    return node_helpers.conditioning_set_values(cond, {"guide_attention_entries": tagged})
+        except Exception:
+            pass
+        return cond
+
+    def _append_guide_latent(self, chunk, guide_frame, apply_at, strength, positive, negative, vae,
+                              phase_tag=0.0):
+        """Append one guide latent frame with LTX guide attention at apply_at.
+
+        `phase_tag`: when non-zero, marks this guide's entry with `funpack_identity_phase` so
+        _install_identity_phase can find its exact token range and apply the source-phase RoPE
+        rotation the Best-FaceID LoRA was trained against — see [[project_identity_transfer_native]].
+        """
         try:
             from comfy_extras.nodes_lt import LTXVAddGuide, _append_guide_attention_entry
         except ImportError:
@@ -3594,6 +3629,9 @@ class FunPackLTXAVSceneChainSampler:
         positive, negative = _append_guide_attention_entry(
             positive, negative, pre_filter_count, guide_latent_shape, strength=float(strength)
         )
+        if phase_tag:
+            positive = self._tag_last_guide_entry(positive, "funpack_identity_phase", float(phase_tag))
+            negative = self._tag_last_guide_entry(negative, "funpack_identity_phase", float(phase_tag))
 
         result = self._clone_latent(chunk)
         tensors = self._latent_tensors(result)
@@ -3807,7 +3845,7 @@ class FunPackLTXAVSceneChainSampler:
             return None
 
     def _append_media_guide_at(self, chunk, filename, frame_idx, apply_at, strength,
-                               positive, negative, vae):
+                               positive, negative, vae, phase_tag=0.0):
         chunk_tensors = self._latent_tensors(chunk)
         if not chunk_tensors:
             return chunk, positive, negative, 0, 0
@@ -3818,13 +3856,15 @@ class FunPackLTXAVSceneChainSampler:
         at = self._resolve_frame_index(total, frame_idx)
         guide_slice = self._time_slice(guide_frame, at, at + 1)
         return self._append_guide_latent(
-            chunk, guide_slice, apply_at, strength, positive, negative, vae,
+            chunk, guide_slice, apply_at, strength, positive, negative, vae, phase_tag=phase_tag,
         ) + (0,)
 
     def _apply_configured_guides(self, chunk, scene_index, guide_list, latent_template,
-                                 scene_outputs, scene_media_by_ref, positive, negative, vae):
+                                 scene_outputs, scene_media_by_ref, positive, negative, vae,
+                                 identity_transfer_enabled=False, identity_transfer_phase=2.0):
         head_crop = 0
         tail_crop = 0
+        identity_phase_applied = False
         for g in guide_list or []:
             if not g or not g.get("enabled", True):
                 continue
@@ -3854,14 +3894,20 @@ class FunPackLTXAVSceneChainSampler:
                 fn = (scene_media_by_ref or {}).get(ref) if ref else None
                 if not fn:
                     continue
+                # The project's Best-FaceID reference face (Continuity's "Identity pin") is the
+                # ONLY guide eligible for the source-phase RoPE tag — never prior-scene/mid-scene
+                # guides, which serve motion continuity rather than a trained-LoRA identity lock.
+                phase_tag = identity_transfer_phase if (identity_transfer_enabled and g.get("identity_pin")) else 0.0
                 chunk, positive, negative, head, tail = self._append_media_guide_at(
-                    chunk, fn, frame_idx, apply_at, strength, positive, negative, vae,
+                    chunk, fn, frame_idx, apply_at, strength, positive, negative, vae, phase_tag=phase_tag,
                 )
+                if phase_tag:
+                    identity_phase_applied = True
             else:
                 continue
             head_crop += head
             tail_crop += tail
-        return chunk, positive, negative, head_crop, tail_crop
+        return chunk, positive, negative, head_crop, tail_crop, identity_phase_applied
 
     def _append_mid_scene_guide(self, chunk, previous_output, positive, negative, vae, strength):
         """Append the middle frame of the previous scene as a guide for the current chunk
@@ -4064,6 +4110,129 @@ class FunPackLTXAVSceneChainSampler:
             except Exception:
                 pass
 
+    # ---------------------------------------------------------------------------
+    # identity_transfer: source-phase RoPE tag for the Best-FaceID-style identity pin.
+    #
+    # FunPack's guide-attention path (LTXVAddGuide.add_keyframe_index + causal_fix, see
+    # _append_guide_latent) already places a reference image on the frame-0 grid with a
+    # tunable attention strength — structurally the same "overlap" placement an identity
+    # LoRA like Best-FaceID expects. The one piece that path doesn't do is the LoRA's
+    # trained-time "source_phase" tag: an extra per-token RoPE rotation that marks those
+    # tokens as coming from a reference image, not the real timeline. This patches that in,
+    # scoped ONLY to the specific guide entry marked funpack_identity_phase (see
+    # _tag_last_guide_entry) — never to co-active guides (mid_scene_guide, carry, prior-
+    # scene continuity) that happen to share the same keyframe_idxs tensor.
+    #
+    # Two bound methods are wrapped on model.model.diffusion_model:
+    #   _prepare_timestep            (receives resolved_guide_entries / num_guide_tokens via
+    #                                 kwargs already computed by the model's own guide-token
+    #                                 accounting) -> locates the tagged entry's exact token
+    #                                 range and stashes it on self for the very next call.
+    #   _prepare_positional_embeddings (runs right after, same forward pass, no kwargs) ->
+    #                                 rotates just that token range's RoPE freqs.
+    # Idempotent + tag/strip per scene (same discipline as _install_v2a_scale /
+    # _strip_funpack_scene_wrappers) so an interrupt mid-scene can never leave a stale patch
+    # steering a later, unrelated run — see [[project_hook_leak_bug]].
+    # ---------------------------------------------------------------------------
+    _IDENTITY_PHASE_TAG = "_funpack_identity_phase_patched"
+
+    def _install_identity_phase(self, model):
+        try:
+            ltxv = model.model.diffusion_model
+        except Exception:
+            return None
+        if getattr(ltxv, self._IDENTITY_PHASE_TAG, False):
+            return None  # already installed (idempotent across scenes/runs)
+
+        orig_prepare_timestep = ltxv._prepare_timestep
+        orig_prepare_pe = ltxv._prepare_positional_embeddings
+
+        def prepare_timestep(self_ltxv, timestep, batch_size, hidden_dtype, **kwargs):
+            self_ltxv._funpack_identity_range = None
+            self_ltxv._funpack_identity_phase_value = 0.0
+            try:
+                entries = kwargs.get("resolved_guide_entries")
+                num_guide_tokens = kwargs.get("num_guide_tokens", 0)
+                if entries and num_guide_tokens:
+                    offset = 0
+                    for entry in entries:
+                        phase = entry.get("funpack_identity_phase")
+                        surviving = int(entry.get("surviving_count", 0))
+                        if phase:
+                            self_ltxv._funpack_identity_range = (offset, surviving, int(num_guide_tokens))
+                            self_ltxv._funpack_identity_phase_value = float(phase)
+                            break
+                        offset += surviving
+            except Exception:
+                self_ltxv._funpack_identity_range = None
+            return orig_prepare_timestep(timestep, batch_size, hidden_dtype, **kwargs)
+
+        def prepare_pe(self_ltxv, pixel_coords, frame_rate, x_dtype):
+            pe = orig_prepare_pe(pixel_coords, frame_rate, x_dtype)
+            rng = getattr(self_ltxv, "_funpack_identity_range", None)
+            phase = getattr(self_ltxv, "_funpack_identity_phase_value", 0.0)
+            if not rng or not phase:
+                return pe
+            try:
+                offset, length, num_guide_tokens = rng
+                # AV models return [(v_pe, av_cross_video), (a_pe, av_cross_audio)]; plain
+                # video models return v_pe directly. Only the video PE carries guide tokens.
+                if isinstance(pe, list) and len(pe) and isinstance(pe[0], (list, tuple)) and isinstance(pe[0][0], (list, tuple)):
+                    v_pe, cross_v = pe[0][0], pe[0][1]
+                    v_pe = self._rotate_identity_phase(v_pe, offset, length, num_guide_tokens, phase)
+                    return [(v_pe, cross_v), pe[1]]
+                return self._rotate_identity_phase(pe, offset, length, num_guide_tokens, phase)
+            except Exception:
+                return pe
+
+        ltxv._prepare_timestep = types.MethodType(prepare_timestep, ltxv)
+        ltxv._prepare_positional_embeddings = types.MethodType(prepare_pe, ltxv)
+        setattr(ltxv, self._IDENTITY_PHASE_TAG, True)
+        return (ltxv, orig_prepare_timestep, orig_prepare_pe)
+
+    def _strip_identity_phase(self, handle):
+        if not handle:
+            return
+        ltxv, orig_prepare_timestep, orig_prepare_pe = handle
+        try:
+            ltxv._prepare_timestep = orig_prepare_timestep
+            ltxv._prepare_positional_embeddings = orig_prepare_pe
+            setattr(ltxv, self._IDENTITY_PHASE_TAG, False)
+            for attr in ("_funpack_identity_range", "_funpack_identity_phase_value"):
+                if hasattr(ltxv, attr):
+                    delattr(ltxv, attr)
+        except Exception:
+            pass
+
+    def _rotate_identity_phase(self, pe, offset, length, num_guide_tokens, phase, theta=10000.0):
+        """Givens-rotate the RoPE freqs of exactly [guide_start+offset, guide_start+offset+length)
+        (a single guide entry's tokens, identified upstream — never the whole guide-token tail)
+        by a phase derived from `phase` — the same source_phase tag the LoRA saw at train time.
+        pe = (cos, sin, split_mode) with cos/sin shaped [..., T, L]. Returns a new pe tuple."""
+        if length <= 0 or not phase:
+            return pe
+        cos, sin = pe[0], pe[1]
+        rest = tuple(pe[2:])
+        T = cos.shape[-2]
+        guide_start = T - int(num_guide_tokens)
+        start = guide_start + int(offset)
+        end = start + int(length)
+        if start < 0 or end > T or start >= end:
+            return pe
+        L = cos.shape[-1]
+        d = torch.arange(L, device=cos.device, dtype=torch.float32)
+        rate = theta ** (-d / float(L))
+        rot = (float(phase) * rate)
+        pc = rot.cos().to(cos.dtype); ps = rot.sin().to(sin.dtype)
+        idx = [slice(None)] * cos.dim()
+        idx[-2] = slice(start, end)
+        idx = tuple(idx)
+        c0, s0 = cos[idx], sin[idx]
+        cos = cos.clone(); sin = sin.clone()
+        cos[idx] = c0 * pc - s0 * ps
+        sin[idx] = s0 * pc + c0 * ps
+        return (cos, sin, *rest)
+
     def _bounded_attention_region_mask(self, t, h, w, device):
         """[T*H*W] region id per video token: 0 = left half (by width), 1 = right half.
         Assumes (t, h, w) row-major flattening — matches the packed-latent layout
@@ -4190,6 +4359,7 @@ class FunPackLTXAVSceneChainSampler:
                output_guidance=False, output_guidance_strength=0.02,
                dynashift=False, dynashift_strength=0.3, dynashift_threshold=0.6,
                alg_guide_blur_strength=2.0, alg_guide_blur_sigma_threshold=0.975,
+               identity_transfer_enabled=False, identity_transfer_phase=2.0,
                unique_id=None, prompt=None):
         if not isinstance(positive, list) or not positive:
             raise ValueError("positive conditioning must contain at least one scene entry.")
@@ -4381,6 +4551,7 @@ class FunPackLTXAVSceneChainSampler:
             soft_carried = 0
             guide_tail = 0
             audio_tail = 0
+            identity_phase_applied = False
             run_mechanisms: list = []
             anchor_meta = (scene_anchors or {}).get(str(scene_index))
             if output is None:
@@ -4390,9 +4561,11 @@ class FunPackLTXAVSceneChainSampler:
                     custom_guides = per_scene_guides[scene_index]
                 if custom_guides:
                     run_mechanisms.append("custom_guide_stack")
-                    chunk, scene_positive, scene_negative, carried, guide_tail = self._apply_configured_guides(
+                    chunk, scene_positive, scene_negative, carried, guide_tail, identity_phase_applied = self._apply_configured_guides(
                         chunk, scene_index, custom_guides, latent_template, scene_outputs, scene_media_by_ref,
                         scene_positive, scene_negative, vae,
+                        identity_transfer_enabled=identity_transfer_enabled,
+                        identity_transfer_phase=identity_transfer_phase,
                     )
                     carried_guide_frames = max(carried_guide_frames, carried)
             elif anchor_meta:
@@ -4432,9 +4605,11 @@ class FunPackLTXAVSceneChainSampler:
                     custom_guides = per_scene_guides[scene_index]
                 if custom_guides:
                     run_mechanisms.append("custom_guide_stack")
-                    chunk, scene_positive, scene_negative, carried, guide_tail = self._apply_configured_guides(
+                    chunk, scene_positive, scene_negative, carried, guide_tail, identity_phase_applied = self._apply_configured_guides(
                         chunk, scene_index, custom_guides, latent_template, scene_outputs, scene_media_by_ref,
                         scene_positive, scene_negative, vae,
+                        identity_transfer_enabled=identity_transfer_enabled,
+                        identity_transfer_phase=identity_transfer_phase,
                     )
                     carried_guide_frames = max(carried_guide_frames, carried)
                 elif carry_i2v_guides:
@@ -4469,6 +4644,7 @@ class FunPackLTXAVSceneChainSampler:
             # mode as the block-hook leak; see _strip_funpack_scene_wrappers).
             _scene_base_wrapper = model.model_options.get("model_function_wrapper")
             _v2a_handles = []
+            _identity_phase_handle = None
             try:
                 if embed_guidance and _value_fn is not None and _value_fn.is_ready():
                     run_mechanisms.append("embed_guidance_vf_ascend")
@@ -4520,6 +4696,10 @@ class FunPackLTXAVSceneChainSampler:
                     _v2a_handles = self._install_v2a_scale(model, v2a_grad_scale)
                     if _v2a_handles:
                         run_mechanisms.append(f"v2a_grad_scale({v2a_grad_scale})")
+                if identity_transfer_enabled and identity_phase_applied:
+                    _identity_phase_handle = self._install_identity_phase(model)
+                    if _identity_phase_handle:
+                        run_mechanisms.append(f"identity_transfer_phase({identity_transfer_phase})")
                 _t_sample0 = _time.perf_counter()
                 sampled = self._sample_chunk(
                     model, sampler, sigmas, scene_seed, cfg, scene_positive, scene_negative, chunk,
@@ -4533,6 +4713,7 @@ class FunPackLTXAVSceneChainSampler:
                 _phase_sampling += _scene_sample_s
             finally:
                 self._remove_v2a_scale(_v2a_handles)
+                self._strip_identity_phase(_identity_phase_handle)
                 if model.model_options.get("model_function_wrapper") is not _scene_base_wrapper:
                     if _scene_base_wrapper is not None:
                         model.model_options["model_function_wrapper"] = _scene_base_wrapper
diff --git a/tests/test_identity_transfer.py b/tests/test_identity_transfer.py
new file mode 100644
index 00000000..6293a76f
--- /dev/null
+++ b/tests/test_identity_transfer.py
@@ -0,0 +1,239 @@
+"""Unit tests for the native Best-FaceID identity_transfer port.
+
+Covers the three pieces added to FunPackLTXAVSceneChainSampler:
+  - _tag_last_guide_entry: best-effort marks the most-recently-appended
+    guide_attention_entry, never touching earlier entries.
+  - _apply_configured_guides: routes the phase tag ONLY to the guide entry the
+    editor marked identity_pin (never prior-scene/mid-scene/template guides that
+    happen to share the same call).
+  - _install_identity_phase / _strip_identity_phase / _rotate_identity_phase: the
+    tag+strip RoPE patch that reproduces the LoRA's trained-time source_phase
+    tag, scoped to the exact token range of the tagged guide (not the whole
+    guide-token tail, which may include co-active unrelated guides).
+"""
+import sys
+import types
+from pathlib import Path
+
+import torch
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+for _name in (
+    "comfy", "comfy.k_diffusion", "comfy.k_diffusion.sampling",
+    "comfy.model_sampling", "comfy.nested_tensor", "comfy.sample",
+    "comfy.samplers", "comfy.utils",
+):
+    sys.modules.setdefault(_name, types.ModuleType(_name))
+sys.modules["comfy.nested_tensor"].NestedTensor = object
+
+import samplers  # noqa: E402
+
+
+def _sampler():
+    return samplers.FunPackLTXAVSceneChainSampler()
+
+
+def _install_node_helpers():
+    """node_helpers.conditioning_set_values, the same real implementation shape
+    _append_guide_attention_entry itself relies on (copy metadata dict, overwrite
+    the given keys) — reimplemented here so the test runs without a full ComfyUI env."""
+    mod = types.ModuleType("node_helpers")
+
+    def conditioning_set_values(conditioning, values=None, append=False):
+        values = values or {}
+        out = []
+        for t in conditioning:
+            n = [t[0], t[1].copy()]
+            for k, v in values.items():
+                n[1][k] = v
+            out.append(n)
+        return out
+
+    mod.conditioning_set_values = conditioning_set_values
+    sys.modules["node_helpers"] = mod
+    return mod
+
+
+def _cond_with_entries(entries):
+    return [[torch.zeros(1, 1, 4), {"guide_attention_entries": list(entries)}]]
+
+
+# ── _tag_last_guide_entry ──────────────────────────────────────────────────────
+
+def test_tag_last_guide_entry_marks_only_newest_entry():
+    _install_node_helpers()
+    s = _sampler()
+    cond = _cond_with_entries([{"pre_filter_count": 4}, {"pre_filter_count": 8}])
+    tagged = s._tag_last_guide_entry(cond, "funpack_identity_phase", 2.0)
+    entries = tagged[0][1]["guide_attention_entries"]
+    assert entries[0].get("funpack_identity_phase") is None
+    assert entries[1]["funpack_identity_phase"] == 2.0
+    # Original untouched (conditioning_set_values copies, doesn't mutate in place).
+    assert "funpack_identity_phase" not in cond[0][1]["guide_attention_entries"][1]
+
+
+def test_tag_last_guide_entry_noop_without_entries():
+    _install_node_helpers()
+    s = _sampler()
+    cond = [[torch.zeros(1, 1, 4), {}]]
+    out = s._tag_last_guide_entry(cond, "funpack_identity_phase", 2.0)
+    assert out is cond  # unchanged object, nothing to tag
+
+
+def test_tag_last_guide_entry_noop_without_node_helpers():
+    sys.modules.pop("node_helpers", None)
+    s = _sampler()
+    cond = _cond_with_entries([{"pre_filter_count": 4}])
+    out = s._tag_last_guide_entry(cond, "funpack_identity_phase", 2.0)
+    assert out is cond
+
+
+# ── _apply_configured_guides phase-tag routing ─────────────────────────────────
+
+def test_phase_tag_applies_only_to_identity_pin_entry(monkeypatch):
+    s = _sampler()
+    calls = []
+
+    def fake_append_media(chunk, filename, frame_idx, apply_at, strength, positive, negative, vae,
+                          phase_tag=0.0):
+        calls.append((filename, phase_tag))
+        return chunk, positive, negative, 0, 0
+
+    monkeypatch.setattr(s, "_append_media_guide_at", fake_append_media)
+    guide_list = [
+        {"enabled": True, "source": "image", "media_ref": "pin", "identity_pin": True, "strength": 0.35},
+        {"enabled": True, "source": "image", "media_ref": "prior", "strength": 0.35},  # NOT the pin
+    ]
+    scene_media_by_ref = {"pin": "pin.png", "prior": "prior.png"}
+    chunk, pos, neg, head, tail, phase_applied = s._apply_configured_guides(
+        {}, 1, guide_list, {}, [], scene_media_by_ref, [], [], object(),
+        identity_transfer_enabled=True, identity_transfer_phase=3.0,
+    )
+    assert calls == [("pin.png", 3.0), ("prior.png", 0.0)]
+    assert phase_applied is True
+
+
+def test_phase_tag_off_when_identity_transfer_disabled(monkeypatch):
+    s = _sampler()
+    calls = []
+    monkeypatch.setattr(s, "_append_media_guide_at",
+                         lambda chunk, filename, *a, phase_tag=0.0, **kw: (calls.append(phase_tag), chunk, [], [], 0, 0)[1:])
+    guide_list = [{"enabled": True, "source": "image", "media_ref": "pin", "identity_pin": True, "strength": 0.35}]
+    _, _, _, _, _, phase_applied = s._apply_configured_guides(
+        {}, 1, guide_list, {}, [], {"pin": "pin.png"}, [], [], object(),
+        identity_transfer_enabled=False, identity_transfer_phase=2.0,
+    )
+    assert calls == [0.0]
+    assert phase_applied is False
+
+
+# ── _install_identity_phase / _strip_identity_phase / _rotate_identity_phase ──
+
+class _FakeLTXV:
+    def _prepare_timestep(self, timestep, batch_size, hidden_dtype, **kwargs):
+        return ("timestep", batch_size, hidden_dtype)
+
+    def _prepare_positional_embeddings(self, pixel_coords, frame_rate, x_dtype):
+        # cos/sin shaped [T=10, L=4]; identity token slice under test is [6:9).
+        T, L = 10, 4
+        cos = torch.arange(T * L, dtype=torch.float32).reshape(T, L)
+        sin = torch.arange(T * L, dtype=torch.float32).reshape(T, L) * -1.0
+        return (cos, sin, False)
+
+
+def _fake_model(ltxv):
+    inner = types.SimpleNamespace(diffusion_model=ltxv)
+    return types.SimpleNamespace(model=inner)
+
+
+def test_install_is_idempotent_and_strip_restores_originals():
+    s = _sampler()
+    ltxv = _FakeLTXV()
+    orig_pt = ltxv._prepare_timestep
+    orig_pe = ltxv._prepare_positional_embeddings
+
+    handle1 = s._install_identity_phase(_fake_model(ltxv))
+    assert handle1 is not None
+    assert ltxv._prepare_timestep is not orig_pt
+    handle2 = s._install_identity_phase(_fake_model(ltxv))
+    assert handle2 is None  # already installed — idempotent, no double-wrap
+
+    s._strip_identity_phase(handle1)
+    # Bound-method objects aren't identity-stable across separate attribute accesses in
+    # CPython (a fresh wrapper is created each time) — compare by equality (same __func__ +
+    # __self__) instead of `is`.
+    assert ltxv._prepare_timestep == orig_pt
+    assert ltxv._prepare_positional_embeddings == orig_pe
+    assert not hasattr(ltxv, "_funpack_identity_range")
+
+
+def test_prepare_timestep_stashes_range_for_tagged_entry_only():
+    s = _sampler()
+    ltxv = _FakeLTXV()
+    handle = s._install_identity_phase(_fake_model(ltxv))
+    entries = [
+        {"pre_filter_count": 16, "surviving_count": 2},                                    # unrelated guide, first
+        {"pre_filter_count": 16, "surviving_count": 3, "funpack_identity_phase": 2.5},      # the tagged one
+        {"pre_filter_count": 16, "surviving_count": 4},                                    # unrelated guide, after
+    ]
+    ltxv._prepare_timestep(torch.zeros(1), 1, torch.float32,
+                           resolved_guide_entries=entries, num_guide_tokens=9)
+    assert ltxv._funpack_identity_range == (2, 3, 9)  # offset past the first entry's 2 tokens
+    assert ltxv._funpack_identity_phase_value == 2.5
+    s._strip_identity_phase(handle)
+
+
+def test_prepare_timestep_clears_range_without_tagged_entry():
+    s = _sampler()
+    ltxv = _FakeLTXV()
+    handle = s._install_identity_phase(_fake_model(ltxv))
+    ltxv._funpack_identity_range = (0, 1, 1)  # stale from a previous scene
+    entries = [{"pre_filter_count": 16, "surviving_count": 5}]
+    ltxv._prepare_timestep(torch.zeros(1), 1, torch.float32,
+                           resolved_guide_entries=entries, num_guide_tokens=5)
+    assert ltxv._funpack_identity_range is None
+    s._strip_identity_phase(handle)
+
+
+def test_prepare_pe_rotates_only_the_tagged_range():
+    s = _sampler()
+    ltxv = _FakeLTXV()
+    handle = s._install_identity_phase(_fake_model(ltxv))
+    baseline_cos, baseline_sin, _ = ltxv._prepare_positional_embeddings(None, 25, torch.float32)
+
+    entries = [
+        {"pre_filter_count": 16, "surviving_count": 2},
+        {"pre_filter_count": 16, "surviving_count": 3, "funpack_identity_phase": 2.5},
+        {"pre_filter_count": 16, "surviving_count": 4},
+    ]
+    ltxv._prepare_timestep(torch.zeros(1), 1, torch.float32,
+                           resolved_guide_entries=entries, num_guide_tokens=9)
+    cos, sin, split_mode = ltxv._prepare_positional_embeddings(None, 25, torch.float32)
+
+    # T=10, num_guide_tokens=9 -> guide_start=1; offset=2, length=3 -> rotated slice [3:6).
+    untouched = list(range(0, 3)) + list(range(6, 10))
+    for i in untouched:
+        assert torch.equal(cos[i], baseline_cos[i]), f"row {i} should be untouched"
+        assert torch.equal(sin[i], baseline_sin[i]), f"row {i} should be untouched"
+    rotated_rows = [3, 4, 5]
+    changed = any(not torch.equal(cos[i], baseline_cos[i]) for i in rotated_rows)
+    assert changed, "rotated slice should differ from baseline"
+    assert split_mode is False  # passthrough element preserved
+    s._strip_identity_phase(handle)
+
+
+def test_rotate_identity_phase_is_noop_for_zero_phase_or_length():
+    s = _sampler()
+    pe = (torch.ones(5, 4), torch.zeros(5, 4), False)
+    assert s._rotate_identity_phase(pe, 0, 0, 5, 2.0) is pe
+    assert s._rotate_identity_phase(pe, 0, 2, 5, 0.0) is pe
+
+
+def test_rotate_identity_phase_rejects_out_of_bounds_range():
+    s = _sampler()
+    cos = torch.arange(20, dtype=torch.float32).reshape(5, 4)
+    sin = -cos.clone()
+    pe = (cos, sin, False)
+    out = s._rotate_identity_phase(pe, offset=10, length=3, num_guide_tokens=5, phase=2.0)
+    assert out is pe  # start/end fall outside T -> safe no-op

From 93dcf8dfc2a0793044d5c652f2a994a3bc003d04 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 20:55:04 +0300
Subject: [PATCH 10/11] Setup-node picker collapses on select; autocomplete
 shows all matching shortcuts
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Models & Pipeline -> +New node: picking a node now collapses the search results
into a one-line "selected" row with a Change button, instead of leaving the full
list open above the Values/Wiring panel.

Prompt autocomplete: dropped the MAX_ITEMS=8 cap on matchTriggers — a prefix
shared by many shortcuts (e.g. 15 "lazy" triggers) now shows all of them; the
menu already scrolls (max-height in CSS), so nothing was gained by truncating.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---
 movie_editor/frontend/autocomplete.js |  6 +++---
 movie_editor/frontend/models.js       | 17 ++++++++++++++++-
 movie_editor/frontend/styles.css      |  3 +++
 3 files changed, 22 insertions(+), 4 deletions(-)

diff --git a/movie_editor/frontend/autocomplete.js b/movie_editor/frontend/autocomplete.js
index 93fa3b6b..abcb2b74 100644
--- a/movie_editor/frontend/autocomplete.js
+++ b/movie_editor/frontend/autocomplete.js
@@ -7,7 +7,6 @@
 
   const DELIMS = /[,\n;]/;          // token boundaries — a trigger lives between delimiters
   const MIN_QUERY = 2;
-  const MAX_ITEMS = 8;
 
   function enabled() { return !!S.getEditorSetting("autocomplete"); }
 
@@ -25,7 +24,9 @@
   }
 
   // Triggers whose text contains `query` (substring, so "@golden" matches "golden"),
-  // prefix matches ranked first, identical triggers de-duped.
+  // prefix matches ranked first, identical triggers de-duped. Unbounded — the menu itself
+  // scrolls (max-height in CSS), so a prefix shared by many shortcuts (e.g. 15 "lazy_*"
+  // triggers) shows every one of them instead of silently hiding the tail.
   function matchTriggers(query) {
     const q = query.toLowerCase();
     const scored = [];
@@ -42,7 +43,6 @@
       const k = e.trigger.toLowerCase();
       if (seen.has(k)) continue;
       seen.add(k); uniq.push(e);
-      if (uniq.length >= MAX_ITEMS) break;
     }
     return uniq;
   }
diff --git a/movie_editor/frontend/models.js b/movie_editor/frontend/models.js
index d1fcc407..d96e2442 100644
--- a/movie_editor/frontend/models.js
+++ b/movie_editor/frontend/models.js
@@ -715,9 +715,23 @@
     setTimeout(() => search.focus(), 0);
 
     let curList = [], curCand = null, draft = null;
+    // Collapsed once a node is picked — the full list is clutter once Values/Wiring are
+    // showing below it. Re-expands on "Change" or as soon as the user edits the search again.
+    let listOpen = true;
 
     function renderList() {
       clear(listBox);
+      if (curCand && !listOpen) {
+        const row = el("div", "ns-row ns-row-selected active");
+        row.append(el("span", "ns-row-name", curCand.display_name));
+        row.append(el("span", "ns-row-class", curCand.class));
+        const change = el("button", "btn ghost tiny ns-change", "Change");
+        change.type = "button";
+        change.onclick = (e) => { e.stopPropagation(); listOpen = true; renderList(); search.focus(); };
+        row.append(change);
+        listBox.append(row);
+        return;
+      }
       const q = search.value.trim().toLowerCase();
       const f = q ? curList.filter((c) => (c.display_name + " " + c.class + " " + (c.category || "")).toLowerCase().includes(q)) : curList;
       if (!f.length) { listBox.append(el("div", "ns-empty", curList.length ? "No match." : "No matching nodes.")); return; }
@@ -730,7 +744,7 @@
       });
       if (f.length > 200) listBox.append(el("div", "ns-empty", `…${f.length - 200} more — refine the search.`));
     }
-    search.oninput = renderList;
+    search.oninput = () => { listOpen = true; renderList(); };
 
     async function pick(cls) {
       curCand = specByClass[cls];
@@ -741,6 +755,7 @@
         wires: { ...(extras.wires || {}) },              // outName -> single target
         input_sources: { ...(extras.input_sources || {}) },
       };
+      listOpen = false;
       renderList(); renderDetail();
     }
 
diff --git a/movie_editor/frontend/styles.css b/movie_editor/frontend/styles.css
index c40a2e40..2caaa071 100644
--- a/movie_editor/frontend/styles.css
+++ b/movie_editor/frontend/styles.css
@@ -1766,6 +1766,9 @@ body.tour-mode { overflow: hidden; }
 .ns-row-name { font-size: 12.5px; color: var(--text); }
 .ns-row-class { font-size: 10.5px; color: var(--faint); font-family: var(--font-mono);
   overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+.ns-row-selected { cursor: default; }
+.ns-row-selected .ns-row-class { flex: 1; }
+.ns-change { margin-left: auto; flex: 0 0 auto; }
 .ns-empty { padding: 10px 12px; font-size: 12px; color: var(--faint); }
 .ns-detail { display: flex; flex-direction: column; gap: 10px; }
 .ns-foot { display: flex; justify-content: flex-end; gap: 8px; padding-top: 4px; }

From 05b628629b899aa7c083c984d96c8beaa31d1970 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 6 Jul 2026 20:59:44 +0300
Subject: [PATCH 11/11] Release 3.2.1: CHANGELOG + version bump

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---
 CHANGELOG.md   | 36 ++++++++++++++++++++++++++++++++++++
 pyproject.toml |  2 +-
 2 files changed, 37 insertions(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index f3a641e8..4e4f3a5b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,42 @@
 
 ## [Unreleased]
 
+## [3.2.1] - 2026-07-06
+
+### Added
+- **Shortcut Ideas bulb** (💡 next to prompt fields) is now a habit-aware co-pilot: alongside
+  the existing missing-category browser, it surfaces shortcuts you've historically paired with
+  what's already in the prompt, and what usually follows the previous scene in your own
+  projects — mined on demand from saved project text, no new persistent state.
+- **Media gallery: one-click continuity pin** — a 📌 hover action on image cards sets/unsets
+  the project's Continuity identity pin directly from the Assets gallery, without a trip to
+  Engine settings.
+- **Shortcut revolver** — an opt-in no-repeat cycling mode for multi-replacement shortcuts
+  (sequential or shuffled). A shortcut with several replacements now draws through all of them
+  before repeating, instead of a seeded random pick that could repeat early.
+- **Guide blur** (`alg_blur_guides`) now stands alone from the sampler's anchor ALG: it works
+  with `alg_enabled` off, and has its own strength / sigma-threshold controls independent of
+  the anchor's.
+- **Best-FaceID identity transfer** (experimental, Engine ▸ Experimental) — tags the existing
+  Continuity "Identity pin" guide with the source-phase RoPE rotation Best-FaceID-style
+  identity LoRAs were trained on. A native port on the Scene Chain Sampler, no external
+  custom-node dependency; the LoRA itself loads the normal way via Models.
+
+### Changed
+- Models & Pipeline "+ New node" search collapses to a compact summary once a node is picked,
+  instead of leaving the full results list open above the Values/Wiring panel.
+- Prompt autocomplete no longer caps suggestions at 8 — a prefix shared by many shortcuts now
+  shows every match (the menu already scrolls).
+
+### Fixed
+- **Preview stability**: the preview-segment ffmpeg re-encode ran synchronously inside the
+  request handler, freezing ComfyUI's event loop and stalling every other streaming response —
+  now backgrounded with per-key locking. Ghost clips (and deep-trimmed lone clips) get proper
+  segments instead of deep-seeking the raw chain file. Segment URLs encode the clip duration so
+  re-trimming a clip can't serve an hour-old cached segment. Pooled preview videos default to
+  `preload=metadata` (only the current clip and its neighbours preload fully), so a full
+  timeline no longer starves the browser's connection pool mid-playback.
+
 ## [3.2.0] - 2026-07-05
 
 ### Added
diff --git a/pyproject.toml b/pyproject.toml
index 8e7f1dc5..1449d9ad 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -2,7 +2,7 @@
 [project]
 name = "ComfyUI-FunPack" # Unique identifier for your node. Immutable after creation.
 description = "An AI movie studio inside ComfyUI: the Cutting Room browser-based video editor with timeline, per-scene prompts and one-click montage generation, plus a self-improving engine that learns from your ratings. Includes the full node stack (Studio, Refiner V2, LTXAV Scene Chain Sampler) for classic graph workflows."
-version = "3.2.0" # Custom Node version. Must be semantically versioned.
+version = "3.2.1" # Custom Node version. Must be semantically versioned.
 license = { text = "GNU GPL v3" }
 dependencies  = [] # Filled in from requirements.txt