Brings stage 1 into stage 2: the gallery drag fix and the change that
follows Anomalous_TTS interface 13 (no server path picker, no storage
change or move, folders shown with the node's settings file).
Conflicts: the picker CSS, the storage-change activity label and its
locale keys are dropped; stage 2's model download and backup entries stay.
tests/tts_setup_api.mjs now checks the pretrained download route instead
of the removed storage and library routes, and that no export sends a
path on the computer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018P2T2R7TFbJsc3C9maCD8B
Anomalous_TTS no longer takes a folder or file path on this computer from
a request (ComfyUI-Manager review of the node): the storage place, other
character folders and pretrained sources are set only in its settings
file, and imported files are always uploaded.
- GPT-SoVITS settings shows the storage place, other places and package
sources without change, move or remove buttons, and names the node's
settings file (status.settings_file) with a copy button and how to edit it.
- The import window drops "from this computer": no server path picker
(ui_tts_path_picker.js is gone), no local folder scan or local clip
preview; a very large upload is confirmed instead.
- tts_setup_api.js loses the removed routes; formatSize moves there.
- The storage-change activity entry, unused locale keys and picker CSS go.
Verified: tools/check_structure.mjs; tts_setup_api in Node with fetch
stubbed; the settings card, batch hero, add menu, file sources and player
rendered with jsdom and stubbed ComfyUI scripts. Not run in ComfyUI.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018P2T2R7TFbJsc3C9maCD8B
Export packs the stores (recipes, combos, materials, parameters, and
optionally ComfyUI's saved workflows), the plugin settings without the
Civitai key, and each model's user layer and own cover into one .zip to
keep on a cloud drive. Import lists what is new, the same or different
and the models it finds again (by SHA-256, else folder, path and size),
adds only what is missing by default, replaces differing files only on
request with this computer's going to the Recycle Bin first, merges a
model's name and notes field by field and puts the user's cover in
place of a Civitai one. Unsafe or foreign zips are refused. Both are
activity entries. tests/test_backup.py covers the rules; testing.md
names a single test file the way unittest finds it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tests/ (Python backend tests, .mjs frontend rule tests) is no longer
ignored. tools/run_tests.mjs runs the structure check, every
tests/*.mjs and the Python tests with $PYTHON, the portable
python_embeded or python. docs/guides/testing.md says how to run and
write them; README links it. Developer notes no longer point at 18 test
files that were removed earlier (AGENTS.md table, backend, frontend and
update-guide notes).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three MCP tools: download_missing_models (Model Check's Download all for
the open workflow, through the page so the placement rules and the put-in
afterwards are the UI's own), download_model (one model from a Civitai
version link, a Hugging Face / GitHub file link or a Civitai hash, into a
folder of its type) and download_status. The default models folder of a
type is shared by the dialog and the tools.
A Civitai version link without a matching file name now takes the
version's primary model file (people rename their files); a hash, when
there is one, still decides.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A model-only LoRA loader was used for every UNet workflow, so a LoRA with text-encoder
weights lost them; and inserting after a checkpoint whose CLIP feeds both prompts was
refused (one link per output only).
- api/mcp_actions.py reads the LoRA's safetensors header: lora_te / te1 / te2 /
text_encoder keys mean it trains the text encoder.
- Such a LoRA becomes a LoraLoader taking MODEL from the model line's end and CLIP from
the same node (checkpoint, full LoRA loader) or a UNet workflow's one CLIP source;
others stay model-only.
- graph_splice.js: spliceAfterOutputs puts a node after outputs on different nodes and
moves every link each output had. The picker's spliceModelChainNode is back to its
committed behavior (the `types` option is gone).
Verified on 8199: UNet + CLIP loader with a text-encoder LoRA, a model-only one after it
and another text-encoder one; a checkpoint feeding two prompts; Ctrl+Z after each.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The second batch of MCP tools: they change things, through the same code the UI uses.
- Canvas, done by the open ComfyUI page: describe_canvas, set_prompt, set_model,
add_lora, place_combo, check_workflow_models, fix_workflow_models, run_workflow,
open_in_anomalous. api/mcp_bridge.py sends the action over ComfyUI's websocket and
waits; web/modules/mcp_bridge.js claims it (visible tab first), does it with
Anomalous's canvas code and answers. Each change is one Ctrl+Z step, shows a toast and
is logged as the AI's (activity entries carry via: "mcp", with their undo).
- This server: scan_models (new / look up again / picked models, via /anomalous/scan_all),
list_voices and speak (Anomalous TTS through /prompt and /history).
- graph_splice.js takes the chain's slot types, so a LoRA can go into a model-only (UNet,
Flux) chain as LoraLoaderModelOnly; the default (MODEL and CLIP) is unchanged.
- Nothing deletes or renames files; acting tools are not marked read-only, so AI apps
ask before calling them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An MCP endpoint (Streamable HTTP) inside ComfyUI's server, so Claude, Cursor, Cherry
Studio and other MCP apps can ask about the user's library without anything extra
installed and without memory or VRAM while nobody is connected.
- api/mcp_server.py: one JSON-RPC message per POST, answered with JSON. Serves the
2026-07-28 stateless protocol (headers must match the body) and the earlier
initialize-based versions; no sessions. This computer only: loopback peer, no proxy
headers, loopback Host and Origin (DNS rebinding).
- api/mcp_tools.py: 11 read-only tools: library overview, model search and details
(optional cover), scan status, output-image search and generation settings
(optional thumbnail), combos, workflow recipes, saved prompts, generated audio,
activity log. Ids resolve inside their roots; tool failures come back as isError.
- docs/guides/mcp.md: setup for each app, what can be asked, privacy note.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Voices has two views switched at the top, Characters and Voice-over, sharing
the character list (clicking a character on Voice-over makes it speak); the
rail's Voices entry reopens the view used last, and the separate Voice-over
rail entry is gone. Generation settings: language as choices and speed as a
slider in view, the sampling numbers folded under fine-tuning with plain
names, no sideways scroll. A finished take says how long it took and links
to the audio gallery. Line cards show editing buttons on hover, the emotion's
reference clip sits beside its chips, and a character with only its main
voice shows no chips. The audio gallery shows each file's lines once and
can open them on Voice-over for another take; the list no longer repeats
the gallery. Also fixes the audio list staying empty after a reload.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Script Director was a drawer behind one of four equal buttons on
Voices, over the character cards, and its name did not say what it does.
It is now Voice-over (配音), a rail page between Voices and Audio
(ui_script_page.js mounts the director); each character card has a
Voice-over button that opens it with that character chosen, and Home's
"Give a character a voice" goes there. Generate is the main button;
pushing, dragging and copying sit under "Into your own workflow", and a
push or drop is one Ctrl+Z step. Plainer words throughout; the Voices
page keeps import, refresh and search.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The README, the help page and the image-side "!" guide still described the
old top tabs and bottom bar and an earlier release's features. They now
cover the rail, Home, model type chips, the activity log, character voices
(with the optional Anomalous_TTS node pack) and the lighter memory use.
- New update guide id 2026-10-rail-and-voices with five steps; the old
guide's locale keys (and the never-shown navigation/transfer/library
ones) are removed.
- Help: the Civitai API key goes in with the key button in the scan wizard,
not in Settings.
- The audio guide says the tag copy button only exists for characters
with emotions.
- update-guide.md loses its section on the removed bottom bar.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The drawer grows to 1000px. From 760px wide the line cards take the full
height and the generation settings, generate bar, result and canvas buttons
sit in a 320px column beside them. When narrow (docked browser) the drawer
covers the studio column and keeps generating below the cards; only the
cards scroll in either layout.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Long scripts squeezed the line cards between the list and a footer that
scrolled on its own. Generation settings (language, speed, sampling) are
now one folded line above the cards that scrolls with them, and the footer
is a short bar: generate / retake / cancel, the result and the canvas
buttons. The fold state is remembered per browser.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The F5-TTS voice library (input/F5-TTS scanning, voice upload with ffmpeg
conversion, romanization, Load Audio clip drags) and the engine switch are
removed. The audio page now works with one engine, Anomalous_TTS
(GPT-SoVITS): characters, drags onto its node, the Script Director with
direct generation, pronunciation and settings. The audio gallery and its
stream route stay (output and temp folders only). The root PATH hack that
existed for F5-TTS's ffmpeg lookup is gone. Recorded as AD-016.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
"Pronunciation" on a character card edits the node's `replace` setting
(script text -> what the character reads). Each row can be heard as the
character reads it now (the saved table applies) and as replaced, through
a Preview Audio run that stays out of the audio gallery; one preview at a
time, cancelled when the dialog closes. The file is re-read before editing
and every other settings field is kept. buildTtsPrompt gains `preview`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Generate section: top_k / top_p / temperature / repetition_penalty folded
under "Advanced", ranges and defaults read from the node's /object_info;
"save as this character's default" keeps language, speed and these (only
values that differ from the node's defaults). The seed is not saved.
- Drawer 560px wide, line cards keep 180px, footer scrolls; for GPT-SoVITS
the bundle preview is hidden and the canvas buttons are compact.
- The audio domain no longer shows the "Audio Workflows" header tab (the
recipe library filtered to "audio"); a `hidden` header button now really
hides despite the header's display !important rule.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Script Director gets a Generate section for GPT-SoVITS characters
(ui_script_run.js): it queues the script as an API prompt with this
page's client id (audio_tts_run.js), shows queued/running/progress,
plays the result and saves it to output/audio/<character>/. The canvas
is never touched. A line card's retake button adds [take:N] to that
line so only it is regenerated; "New take of all" changes the seed.
Cancel deletes a queued job or interrupts it by prompt id when running
(checked against /queue, since the start event can arrive first).
Events that arrive before /prompt answers are buffered and replayed.
Language/speed start from the character's `defaults` and can be saved
back (read-modify-write, unknown fields kept). Needs Anomalous_TTS
interface 11; older versions get a hint and keep push-to-node.
The Arona multi-voice template read a file only one machine has; the
button, route, helper (alignedVoiceValue) and strings are removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A folder holding one character's GPT + SoVITS weights and its clips is that
character, named after the folder: two characters whose weights are both
"model" no longer merge, pinyin weights in a Chinese-named folder take the
folder's name, version folders read "派蒙 v2". Package weight folders keep
the name-by-stem rule.
- Skip rules match the node (interface v9): Python environments and base
models always; GPT_SoVITS / logs / output / TEMP / tools only inside a
package; the chosen or dropped folder never. The drop walker and the folder
dialog apply them where the listing is known (the old re-check on each
file's path skipped a chosen folder called "output"), local scans prefix
the chosen folder so every source sorts alike, and the window names the
skipped and too-deep folders instead of losing files quietly. Drops only
open importable files and stop at 5000.
- Same-name files inside a character are renamed with their folder
(难过_01.wav, sent as the file's "name") instead of dropped; an unclaimed
.list goes to every character.
- Big adds open 20 characters (about 800 files) at a time; "next batch" opens
the rest, and it opens by itself once a batch is imported, replacing the
imported cards. Unassigned files upload only once put into a character.
- Every clip can be previewed while importing, local paths included (through
the node's preview route): one at a time, ▶ turns into stop, another clip
switches.
- Upload queue and preview player move to their own modules, the "add more"
menu to the screens module (ui_tts_import.js 647 → 652 lines with batches and renames added).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012RMuwbsWZHnx2eCWpxYqAB
- A folder per character: clips and line files follow the weights in their
nearest folder, so pinyin weights, Chinese folders and English clips still
land together (a real 34-character folder: 3065 unassigned files -> 0).
- Clips outside 3-10 s and their same-name line files are left out of the
import and folded under "left out" instead of one warning per clip.
- GPT-SoVITS program and training folders are skipped in drops (not even
walked) and picked folders; a very large upload (>1500 files or 2 GB)
first offers the local folder picker.
- Uploads run three at a time, status repaints are coalesced, a card shows
its first 60 clips, move menus are filled when opened, and the unassigned
card folds into one line per folder that can be assigned at once.
- New ui_tts_import_sources.js brings files in (dialogs, drops, local picker).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The import window now asks what the user has first: one character, many
characters or a package, or files for an existing character (picked from
a list, or opened straight from a card's "Add files"). Dropped files skip
the question.
Each character card leads with a checklist (GPT model, SoVITS model,
clips, main voice, name) that lights up the next step and offers its
button in that line; the chosen weights sit in their lines, and the
card header says how many steps are left. A main voice (3-10 s) is now
required for a new character, as GPT-SoVITS needs one.
A spotlight tour explains the first question and the card the first time
each is shown; "?" in the header replays it. ui_spotlight_tour.js takes
its steps and an onClose from the caller now; the browser tour is
unchanged. The fixed screens and tour steps live in
ui_tts_import_screens.js.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The two-column workbench was hard to read on first open. The window now
opens as one drop area that says what to bring (GPT model, SoVITS model,
3-10 s clips) with two buttons and a quiet "files on this PC" line, and
three steps at the top show where the user is.
With files, each character is a card: folded, one line with the name
(edited in place), a short status and what it holds; unfolded, two model
tiles, one line per clip (main-voice star, emotion, line) and the line
files and language at the bottom. One card is open at a time; files that
match no character get a card of their own at the top. "Add more" is a
single button with a menu, the owner picker on each clip only shows on
hover, and the footer says what still needs a look and where it saves.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The import form becomes a two-column workbench for one or many
characters. Files and whole folders can be dropped, or a folder on this
PC scanned (Anomalous_TTS browse?recursive=1, interface 6). Weights pair
up by name with the latest epoch kept; clips and line files follow their
folder or file-name prefix; anything unclear waits in an unassigned
tray. The left column lists the characters with a status dot, the right
shows the selected one with one compact line per clip and a picker (or
drag handle) to move files between characters. Ready characters are
imported one after another, each on its own, so a failure leaves the
others alone.
Grouping rules live in tts_import_groups.js, the draft pane in
ui_tts_import_draft.js and OS drop reading in ui_tts_file_drop.js.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each section title gets a mark (green check with counts, amber
"required" for what a new character still needs, red on problems) and
each row a coloured bar: green ready or merged, blue uploading, grey
skipped, red failed or refused. Clips scroll in their own box and the
Import button stays in view.
When adding to a character, inspect now sends the target (Anomalous_TTS
interface 5): files it already has show as skipped, annotation files as
merged, and a different file under the same name blocks the import.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The sidebar groups characters by language (GPT-SoVITS) or folder
(F5-TTS) in foldable groups; a character unfolds into its clips and a
click plays one through the studio row. Characters drag onto nodes from
the sidebar, a red dot marks characters that need a look, a search box
shows from six characters on, and files dropped on a GPT-SoVITS
character open "Add files" for it.
GPT-SoVITS settings move to a dialog opened from the sidebar footer,
which carries the reminder dot; "Import character" moves into the
studio toolbar. Canvas drag helpers move to audio_voice_drag.js so the
sidebar and the cards share them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The setup card is one folded line; missing pretrained files (fetched on
first use anyway) only get a small dismissable dot once the studio's
characters need them, and the import form notes what the chosen
language will download. An empty GPT-SoVITS studio is a drop area with
the import button.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Uses Anomalous_TTS interface v4.
- Setup card: "where characters are kept" replaces the library list.
"Change..." picks a folder on the ComfyUI machine and asks whether the
current characters move along (move / don't move / cancel). A move shows
its progress (characters, and bytes when copying across disks) and the
studio redraws when it ends; an error says where it stopped.
- "Other places" lists earlier storage places that still hold characters
and yaml folders; x on an earlier place stops listing it (files kept).
- Import form: no location picker; new characters go to the storage place
and the form shows the exact path.
- The "add library" action is gone (no longer offered by the node).
Tested: node tests/tts_setup_api.mjs (storage calls, moving summary).
Isolated ComfyUI in the browser: change storage with move (both
characters moved, a non-character folder left, config updated), change
back without moving (old place listed under other places, stale move
message not shown), remove the earlier place (characters unlisted, files
kept), import form shows the storage path; a .list dropped on a clip's
text box fills it without being added as a file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Each clip shows where its line came from (same-name .txt / .lab,
annotation file, or the file name, marked as a guess to check).
- Dropping a .txt / .lab / .list on a clip's line box fills it: a plain
file is the line, an annotation file is searched for that clip (same
rule as the node, including the name without the emotion part).
- .lab is an importable line file.
Tested: node tests/tts_setup_api.mjs (textFromFile added).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Fixed sections for GPT weight, SoVITS weight, reference clips and line
files; files dropped anywhere on the form go to their section. One GPT
and one SoVITS: a new one replaces the old, and from a batch the latest
epoch is kept (the others are named in a note).
- The save location is one line ("Saved to …/gpt_sovits/<name>") with
"Change location"; the library list only shows when asked for.
- Name clashes are shown while typing; an exact match offers "add to that
character instead", which switches the form to add mode in place.
- Unsupported SoVITS versions (v3/v4) are marked and block the import;
clips outside 3-10 s cannot be the main voice.
- After an import the studio scrolls to the new card and highlights it.
Tested: node tests/tts_setup_api.mjs (epoch pick, name clash, version
block, clip range added). Isolated ComfyUI in the browser: name clash and
switch to add mode, three GPT epochs -> e15 kept, SoVITS replaced by a v4
file -> blocked, short clips not selectable as main, add-mode import of a
clip as {平静} with the card highlighted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Uses Anomalous_TTS interface v3 so a new user can get going without
editing extra_model_paths.yaml or sorting model folders by hand.
- Setup card on the GPT-SoVITS studio: character libraries (add/remove),
pretrained files (per-item or all download with progress, or use an
existing GPT-SoVITS package) and missing Python packages with the
install command. Collapses to one line when ready.
- Import form: drop files, pick them in the browser, or pick them on the
ComfyUI machine (server-side path picker, no upload). Name, SoVITS
version, clip length, lines, main voice and language are filled in from
the node's inspect; emotions per clip. Uploads go in 8 MB chunks.
- "Add files" on a card, or a drop onto it, adds clips and emotions to
that character without replacing its main voice.
- Hidden for nodes without v3; disabled when the page is not opened on
the ComfyUI machine.
- New modules: tts_setup_api.js (calls + pure form rules),
ui_tts_setup.js, ui_tts_import.js, ui_tts_path_picker.js and
styles/12-tts-setup.css. ui_audio_studio.js only wires them in.
Tested: node tests/tts_setup_api.mjs (new, local) plus the audio engine,
node target and script rule tests. Isolated ComfyUI with both plugins in
the browser: setup card for a fresh user, package as pretrained source,
library add, import from local paths (lines found in all.txt), file drop
onto a card (chunked upload, add mode), cancel discards the upload.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Cache engine presence and voice groups (60 s) so domain switches, sidebar
clicks and the studio + sidebar pair cost no requests; saves and added
voices invalidate, Refresh also asks Anomalous_TTS to rescan (?refresh=1).
- Follow Anomalous_TTS interface v2: the list is a summary; the emotion
editor opens at once and loads the character's audio list in the
background (save waits for it, keeps unknown settings fields).
- Cards: show "角色/版本" as name + quiet suffix, one-line drag hint,
header wraps instead of squeezing.
- Docs: ARCHITECTURE audio_engines paragraph, audio-studio guide.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q6FStZ5hiKELvNH3WKgBnx
- audio_engines.js: F5-TTS and GPT-SoVITS (the separate Anomalous_TTS node
pack), detected through /object_info/<node class>. Both engines' voices are
normalised into one voice-group shape so cards, sidebar and Script Director
share one implementation.
- The audio page shows an engine switch; a missing engine stays listed as
"not installed" with install steps (no popups). The F5-TTS library still
works without the F5-TTS node; GPT-SoVITS is never scanned when missing.
- GPT-SoVITS cards come only from the node's HTTP contract: preview reference
clips, drag a character onto AnomalousTTS_CharacterSpeech, and edit the
emotion references ("Edit emotions"), saved through /anomalous_tts/settings
with unknown fields preserved and the node's validation messages shown.
- Drop targets and the Script Director are per engine: voices of one engine
are refused on the other engine's node with the reason.
- Audio gallery reads the character of GPT-SoVITS generations.
- Usage guide (in-app and docs/guides/audio-studio.md) covers both engines.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W411Z69FC3em9CzpeQEub5
- audio_node_targets.js is the only list of nodes the audio studio writes
into: F5-TTS (Audio / Advanced / FromModel) takes a whole character, so the
card header sets its main voice; ComfyUI's Load Audio takes one exact clip,
so a row sets that file. Unlisted nodes are refused instead of matched by
widget name, and a single emotion clip is never written into F5-TTS's
sample (it would break {emotion} tags).
- While dragging, the hint says what releasing will do, or why a hovered node
is refused and what to do instead (shared bindMaterialDrag targetHint /
rejectHint).
- Script Director uses the same table and refuses nodes whose sample_audio is
linked.
- In the audio domain the header "!" opens a five-step usage guide;
docs/guides/audio-studio.md is the full guide (docs/guides is now tracked).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W411Z69FC3em9CzpeQEub5
