diff --git a/CHANGELOG.md b/CHANGELOG.md
index afd87599..7be2d2fd 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,7 +4,7 @@ What is new in each version, in plain words. Newest first.
 
 ## [Unreleased]
 
-- **MiniMax H3-Image** (experimental): a new "Start from" pipeline, MiniMax H3-Image · Text to Image and Edit. Write a prompt for a picture, or add reference pictures and refer to them as `<Picture 1>`, `<Picture 2>`, ... in the prompt (the first is the picture to edit; up to 16). The picture is held for the clip's length like other image models. You need the model, its VAE and its text encoder files; the model's licence excludes the EU, the UK, South Korea and the United States.
+- **MiniMax H3-Image** (experimental): a new "Start from" pipeline, MiniMax H3-Image · Text to Image and Edit. Write a prompt for a picture, or add reference pictures and refer to them as `<Picture 1>`, `<Picture 2>`, ... in the prompt (the first is the picture to edit; up to 16). The result is a PNG picture, not a clip, and Length and FPS are not offered for it (both are set to 1). You need the model, its VAE and its text encoder files; the model's licence excludes the EU, the UK, South Korea and the United States.
 - **Noise merging is now called Step merging**, because most of its modes merge the model's answers at a step, not noise. Settings are kept.
 - **Alternating prompts** (Settings > general): write `[dance|zoom]` in a prompt, or `DANCE|ZOOM` between two shortcuts, for two options, A and B. "Main prompt reads" sets what the prompt says by default: "A while B" (default), "A B", only A or only B. Switch on "Read A and B separately" (off by default) to also assign steps: "Steps that read only A" and "Steps that read only B" give each step you name its own option, with no extra model pass. A `|` next to anything that is not a shortcut is left as typed. Two options per bracket.
 - **Simultaneous actions** (Step merging, needs "Read A and B separately"): with `[A|B]` in the prompt, each merged step is answered once on A and once on B from the same picture, and the two answers are mixed at your strength. "Second pass new noise" sets how much of the noise B starts from is new (0 keeps the same noise for both). Strength 0 reads only A, and strength 1 with new noise 1 reads only B, each with a single pass. Steps named in Step merging win over the per-step lists. Costs one extra model pass per merged step.
diff --git a/app/ui/features/inspector/scene.js b/app/ui/features/inspector/scene.js
index 737e6390..b75a24d7 100644
--- a/app/ui/features/inspector/scene.js
+++ b/app/ui/features/inspector/scene.js
@@ -2,7 +2,7 @@
 import { composer as c } from "../../composer/composer.js";
 import { namesOf, pick } from "./mediapick.js";
 import { effFrames, effFps, genUnitId, isSubclip, seconds, snapFrames } from "../../shell/scenes.js";
-import { gridOf } from "../../shell/frame_grid.js";
+import { gridOf, offersLength } from "../../shell/frame_grid.js";
 import { restoreToPlan } from "../../shell/edits.js";
 
 const MODES = [{ value: "project", label: "Project default" }, { value: "timeline", label: "Timeline trim" }, { value: "custom", label: "Custom" }];
@@ -37,7 +37,7 @@ function drift(sc, open) {
 export const card = (...children) => c.card.default({ children: children.flat().filter(Boolean) });
 
 export function sceneRows(p, app) {
-  const sc = p.selected, open = p.project;
+  const sc = p.selected, open = p.project, slots = app.pipeline && (app.pipeline.view || app.pipeline.slots)();
   if (!sc) return [c.emptyState.default({ icon: "▭", title: "No scene", hint: "Add one on the timeline." })];
   const cuts = p.scenes.filter((s) => genUnitId(s) === genUnitId(sc)).length > 1;
   return [
@@ -49,13 +49,14 @@ export function sceneRows(p, app) {
     card(mediaField(app, p, sc, { label: "Aspect ratio source", hint: "The scene's picture. Its shape sets the aspect ratio; the project's Width and Height set the size.",
       key: "source_image", multiple: false, kinds: ["image", "video"], empty: "— choose aspect ratio source —" }),
     mediaField(app, p, sc, { label: "References", key: "references", multiple: true, kinds: ["image", "video", "audio"], empty: "+ Add reference" })),
-    c.label.section({ text: "Length" }),
-    card(c.field.row({ fields: [
-      lengthField(p, sc, "Frames", "frames_mode", "frames", effFrames(sc, open), gridOf(app.pipeline && (app.pipeline.view || app.pipeline.slots)())),
-      lengthField(p, sc, "FPS", "fps_mode", "fps", effFps(sc, open)),
-    ] }),
-    drift(sc, open),
-    c.button.md({ label: "Generate this scene", tone: "primary", onClick: () => app.say("generate.scene") })),
+    ...(offersLength(slots) ? [c.label.section({ text: "Length" }),
+      card(c.field.row({ fields: [
+        lengthField(p, sc, "Frames", "frames_mode", "frames", effFrames(sc, open), gridOf(slots)),
+        lengthField(p, sc, "FPS", "fps_mode", "fps", effFps(sc, open)),
+      ] }),
+      drift(sc, open),
+      c.button.md({ label: "Generate this scene", tone: "primary", onClick: () => app.say("generate.scene") }))] : [
+      card(c.button.md({ label: "Generate this scene", tone: "primary", onClick: () => app.say("generate.scene") }))]),
     card(c.checkboxRow.default({ label: "Exclude from full generation", hint: "Generate leaves this scene out.", checked: Boolean(sc.excluded), onChange: (v) => p.setScene(sc.id, "excluded", v) })),
     c.banner.info({ text: "Made with the project's pipeline and Engine settings", action: { label: "Engine settings →", onClick: () => app.openSettings && app.openSettings("engine") } }),
   ];
diff --git a/app/ui/features/settings/models.js b/app/ui/features/settings/models.js
index 3d8fbb04..f87e979c 100644
--- a/app/ui/features/settings/models.js
+++ b/app/ui/features/settings/models.js
@@ -228,7 +228,10 @@ export const models = (app) => function mount() {
     const pick = found.find((p) => p.id === id);
     if (!pick || !(await confirm("Replace pipeline", `Replace the whole pipeline with “${pick.title}”? Every file and value you set here is cleared. Revert puts it back until you leave this page.`, "danger"))) return;
     const r = await ps.restore({ slots: pick.slots, removed: [], unwired: {} });
-    if (!r.refused.length) Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v));      // project settings the preset goes with (render steps)
+    if (!r.refused.length) {      // project settings the preset goes with (render steps, a picture model's one frame)
+      Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v));
+      Object.entries(pick.project || {}).forEach(([k, v]) => app.project.setField(k, v));
+    }
     specs = {}; await describe(); say(r.refused.length ? `Not changed: ${r.refused[0]}` : "");
   }
   async function importWorkflow() {
diff --git a/app/ui/features/wizard/index.js b/app/ui/features/wizard/index.js
index 0343fd09..2de0d857 100644
--- a/app/ui/features/wizard/index.js
+++ b/app/ui/features/wizard/index.js
@@ -28,7 +28,7 @@ export default {
       const pick = presets.find((p) => p.id === ctx.preset);
       try {
         await app.project.newProject((ctx.name || "").trim() || "Untitled montage");
-        if (pick) { const r = await app.pipeline.restore({ slots: pick.slots, removed: [], unwired: {} }); if (r.refused.length) c.toast.warn({ text: `The model's pipeline was not loaded: ${r.refused[0]}` }); else Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v)); }
+        if (pick) { const r = await app.pipeline.restore({ slots: pick.slots, removed: [], unwired: {} }); if (r.refused.length) c.toast.warn({ text: `The model's pipeline was not loaded: ${r.refused[0]}` }); else { Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v)); Object.entries(pick.project || {}).forEach(([k, v]) => app.project.setField(k, v)); } }
       } catch (err) { busy = false; return c.toast.warn({ text: `Could not make the project: ${err.message}` }); }
       busy = false;
       close();
diff --git a/app/ui/shell/frame_grid.js b/app/ui/shell/frame_grid.js
index 6bee9470..1e1c3b1b 100644
--- a/app/ui/shell/frame_grid.js
+++ b/app/ui/shell/frame_grid.js
@@ -21,3 +21,7 @@ export async function learnGrid(api, slots) {
   try { Object.assign(specs, (await api.describeNodes(missing)).nodes || {}); } catch { missing.forEach((n) => { specs[n] = null; }); }
   return true;
 }
+
+/** Whether the pipeline takes a length or a rate from the project at all. A model that makes a picture holds it for a fixed time, so
+ *  there is nothing to set: the Length and FPS controls are not offered. A pipeline not known yet offers them. */
+export const offersLength = (slots) => !(slots && slots.length) || slots.some((s) => (s.roles || []).some((r) => r.drives === "frames" || r.drives === "fps"));
diff --git a/app/ui/shell/frame_grid.test.js b/app/ui/shell/frame_grid.test.js
index a35c5980..86469e1e 100644
--- a/app/ui/shell/frame_grid.test.js
+++ b/app/ui/shell/frame_grid.test.js
@@ -1,6 +1,6 @@
 import test from "node:test";
 import assert from "node:assert/strict";
-import { gridOf, learnGrid } from "./frame_grid.js";
+import { gridOf, learnGrid, offersLength } from "./frame_grid.js";
 
 test("the frame grid is read from the model's own length input: H3 17k+5, LTX 8k+1, a plain count none", async () => {
   const slots = [{ id: "r2v", node: "H3Node", roles: [{ at: "project.video", input: "length", drives: "frames" }] },
@@ -14,3 +14,14 @@ test("the frame grid is read from the model's own length input: H3 17k+5, LTX 8k
   await learnGrid({ describeNodes: async () => ({ nodes: { LtxNode: { widgets: [{ name: "length", min: 1, step: 8 }] } } }) }, ltx);
   assert.deepEqual(gridOf(ltx), { step: 8, base: 1, fps: null });
 });
+
+test("Length and FPS are offered when a pipeline takes them, or is not known yet, and not for one that holds a picture for a fixed time", () => {
+  const length = { id: "r", roles: [{ at: "project.video", input: "length", drives: "frames" }] };
+  const rate = { id: "s", roles: [{ at: "project.video", input: "fps", drives: "fps" }] };
+  const size = { id: "a", roles: [{ at: "project.video", input: "width" }] };
+  assert.equal(offersLength([size, length]), true);
+  assert.equal(offersLength([size, rate]), true);
+  assert.equal(offersLength([size, { id: "x" }]), false);
+  assert.equal(offersLength([]), true);
+  assert.equal(offersLength(undefined), true);
+});
diff --git a/core/routes.py b/core/routes.py
index 8f5b7e0f..56b169f6 100644
--- a/core/routes.py
+++ b/core/routes.py
@@ -328,7 +328,9 @@ def register(routes, prefix=None):
                 if isinstance(preset, dict) and preset.get("id") and preset.get("slots"):
                     found.append({"id": preset["id"], "title": preset.get("title") or preset["id"],
                                   "module": spec.id, "slots": preset["slots"],
-                                  "prefs": preset.get("prefs") if isinstance(preset.get("prefs"), dict) else {}})
+                                  "prefs": preset.get("prefs") if isinstance(preset.get("prefs"), dict) else {},
+                                  # project numbers the pipeline goes with (a picture model: 1 frame at 1 fps)
+                                  "project": {k: v for k, v in (preset.get("project") or {}).items() if k in ("num_frames_per_scene", "frame_rate") and isinstance(v, int)}})
         return web.json_response({"presets": found})
 
     @routes.post(P + "/api/pipeline/import")
diff --git a/core/tests/test_routes_pipeline.py b/core/tests/test_routes_pipeline.py
index 70f876d2..567b01a7 100644
--- a/core/tests/test_routes_pipeline.py
+++ b/core/tests/test_routes_pipeline.py
@@ -532,6 +532,13 @@ def test_a_module_provided_preset_is_offered(server, registered):
     assert preset["slots"]
 
 
+def test_a_picture_model_preset_carries_the_project_numbers_it_goes_with(server, registered):
+    _status, body = _request(server, "GET", "/fun/api/pipeline/presets")
+    image = next(p for p in body["presets"] if p["id"] == "minimax_h3_image")
+    assert image["project"] == {"num_frames_per_scene": 1, "frame_rate": 1}
+    assert next(p for p in body["presets"] if p["id"] == "minimax_h3_reference_to_video")["project"] == {}      # a video model changes none
+
+
 def test_a_comfyui_workflow_imports_into_slots_that_build_a_real_graph(server, registered):
     """Over HTTP, against ComfyUI's own nodes: the API export of a plain text-to-image graph."""
     from core import graph
diff --git a/docs/how-to-start.md b/docs/how-to-start.md
index 3b021df9..af18b1a2 100644
--- a/docs/how-to-start.md
+++ b/docs/how-to-start.md
@@ -24,7 +24,7 @@ Start ComfyUI as usual and open `/fun/` on the same address, for example `http:/
 |---|---|
 | MiniMax H3 · Reference to Video, MiniMax H3 · Fast & Simple R2V | The model FunUI is built and tested around. |
 | Stable Diffusion 1.5 · Text to Image (held as a clip) | Experimental. |
-| MiniMax H3-Image · Text to Image and Edit (held as a clip) | Experimental. Add pictures as references: the first is the picture to edit (`<Picture 1>` in the prompt), the others are references (`<Picture 2>`, ...), up to 16. Uses 6 steps at CFG 1; Width and Height set the size (about 4 megapixels is a good start). |
+| MiniMax H3-Image · Text to Image and Edit | Experimental. Add pictures as references: the first is the picture to edit (`<Picture 1>` in the prompt), the others are references (`<Picture 2>`, ...), up to 16. The result is a picture file (PNG), and Length and FPS are not offered (both are set to 1). Uses 6 steps at CFG 1; Width and Height set the size (about 4 megapixels is a good start). |
 | LTX 2.3 · Image to Video, from References, LTX 2.3 · Video to Video, direct | Tested on a real model, with LoRAs, Step merging and Attention explorer. Other enhancements are not tested on LTX yet; please report problems in [Issues](https://github.com/digital-garbage/FunUI/issues). |
 
 ## 4. Pick your files
diff --git a/modules/models/minimax_h3_image/__init__.py b/modules/models/minimax_h3_image/__init__.py
index e312eb8c..bca3c8a8 100644
--- a/modules/models/minimax_h3_image/__init__.py
+++ b/modules/models/minimax_h3_image/__init__.py
@@ -5,9 +5,8 @@ the 16x latent grid, has no sound stream and reads at most 16 pictures: the firs
 picture to edit (`<Picture 1>` in the prompt), the others are references (`<Picture 2>`, ...).
 It samples in 6 steps at CFG 1 and works best around 4 to 16 megapixels.
 
-Like every image model here, its picture is held for the clip's length, so the editor gets a clip
-made the same way as any other. A held 16-megapixel frame is about 48 MB in memory: keep the
-length short at large sizes.
+The result is a picture file (PNG), not a clip. The pipeline offers no Length or FPS; the preset sets
+the project's Length and FPS to 1, so a scene made with it counts as one second on the timeline.
 
 The checkpoint is recognised by its header: the packed-DiT keys of H3 with no audio projection.
 Everything the model needs at run time (the image latent, the picture labels, the sigma grid
@@ -31,6 +30,9 @@ _ABSENT = "audio_patch_proj.weight"
 #: How many pictures the model reads (the one to edit and its references).
 MAX_REFERENCES = 16
 
+#: The project numbers the preset sets: one frame at one frame per second.
+PROJECT = {"num_frames_per_scene": 1, "frame_rate": 1}
+
 
 def detect(keys) -> bool:
     keyset = set(keys)
@@ -52,7 +54,6 @@ def probe_traits(keys) -> list:
 
 
 def text_to_image():
-    video = {"at": "project.video", "input": "fps", "label": "FPS", "drives": "fps"}
     slots = [
         {"id": "model", "group": "Loaders", "node": "FunUIDiffusionModelLoader", "inputs": {
             "weight_dtype": "default", "compute_dtype": "default", "attention": "default"}},
@@ -94,11 +95,8 @@ def text_to_image():
 
         {"id": "decode", "group": "Render", "node": "FunUIDecode",
          "inputs": {"samples": ["sampler", 0], "vae": ["vae", 0], "model": ["model", 0]}},
-        {"id": "hold", "group": "Render", "node": "RepeatImageBatch",
-         "roles": [{"at": "project.video", "input": "amount", "label": "Length", "drives": "frames"}],
-         "inputs": {"image": ["decode", 0], "amount": 1}},
-        {"id": "save", "group": "Render", "node": "FunUISaveVideo", "roles": [video],
-         "inputs": {"images": ["hold", 0], "fps": 24.0, "filename_prefix": "FunUI", "settings": ["settings", 0]}},
+        {"id": "save", "group": "Render", "node": "SaveImage",
+         "inputs": {"images": ["decode", 0], "filename_prefix": "FunUI"}},
     ]
     for n in range(1, MAX_REFERENCES + 1):
         # Not wired here: an unused picture input is simply absent. The app wires the ones a scene has, per run.
@@ -111,8 +109,8 @@ def text_to_image():
 
 
 def presets():
-    return [{"id": "minimax_h3_image", "title": "MiniMax H3-Image · Text to Image and Edit (held as a clip)",
-             "slots": text_to_image()}]
+    return [{"id": "minimax_h3_image", "title": "MiniMax H3-Image · Text to Image and Edit",
+             "slots": text_to_image(), "project": PROJECT}]
 
 
 PROVIDES = {"detect": detect, "probe_traits": probe_traits, "pipeline_presets": presets}
diff --git a/modules/models/minimax_h3_image/tests/test_h3_image.py b/modules/models/minimax_h3_image/tests/test_h3_image.py
index 61ea8d4c..5054be5e 100644
--- a/modules/models/minimax_h3_image/tests/test_h3_image.py
+++ b/modules/models/minimax_h3_image/tests/test_h3_image.py
@@ -54,3 +54,17 @@ def test_the_preset_builds_a_real_graph_and_is_offered(registered):
     assert refs[0]["roles"][0]["wireTo"] == {"slot": "cond", "input": "images.image_1"}     # image_1 is the picture to edit
     sampler = next(s for s in slots if s["id"] == "sampler")["inputs"]
     assert (sampler["steps"], sampler["cfg"], sampler["sampler_name"], sampler["scheduler"]) == (6, 1.0, "euler", "simple")
+
+
+def test_it_saves_a_picture_and_offers_no_length_or_fps(registered):
+    from core import graph, routes
+    from modules.models.minimax_h3_image import PROJECT, presets, text_to_image
+
+    slots = {s["id"]: s for s in text_to_image()}
+    assert not [r for s in slots.values() for r in s.get("roles") or [] if r.get("drives") in ("frames", "fps")]
+    assert slots["save"]["node"] == "SaveImage" and "hold" not in slots          # a PNG, not a clip
+    assert PROJECT == {"num_frames_per_scene": 1, "frame_rate": 1} and presets()[0]["project"] == PROJECT
+    prompt, _problems = graph.build(list(slots.values()))
+    assert [n["class_type"] for n in prompt.values()].count("SaveImage") == 1
+    served = [p for _s, make in routes.modules().providers("pipeline_presets") for p in make() if p["id"] == "minimax_h3_image"]
+    assert served and served[0]["project"] == PROJECT