MTP changes #196

New issue

Closed

opened 2026-10-26 22:10:34 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 22:10:34 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/embedded-docs/pull/196
Author: @lin-bot23
Created: 10/21/2026
Status: ✅ Merged
Merged: 10/24/2026
Merged by: @comfyui-wiki

Base: main ← Head: daily-docs-update-1022

📝 Commits (8)

8f68cf0 Add docs for Mirelo, VEED, Pad Image and Fibo 2.0 nodes
64390f5 docs: follow the Fibo 2.0 type, Krea ultra mode, Bria expand limits and HeyGen Voice pricing
36af12b docs: add 14-language translations for the new nodes
f8bcdf8 docs: add the fibo2 encoder type to CLIP Loader
dbe3567 docs: add TextEncodeMOSSImageEdit and follow the MOSS edit, Fibo 2.0 negative prompt and MTP changes
0f695ab docs: add 14-language translations for TextEncodeMOSSImageEdit and the changed rows
bd357ff Merge origin/main into daily-docs-update-1022
329e0ed Merge origin/main; update TextEncodeMOSSImageEdit for ComfyUI #17320

📊 Changes

241 files changed (+4309 additions, -211 deletions)

View changed files

📝 comfyui_embedded_docs/docs/BriaExpandImage/ar.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/de.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/en.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/es.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/fa.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/fr.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/he.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/it.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/ja.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/ko.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/pt-BR.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/ru.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/tr.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/zh-TW.md (+3 -3)
📝 comfyui_embedded_docs/docs/BriaExpandImage/zh.md (+3 -3)
📝 comfyui_embedded_docs/docs/CLIPLoader/ar.md (+3 -2)
📝 comfyui_embedded_docs/docs/CLIPLoader/de.md (+3 -2)
📝 comfyui_embedded_docs/docs/CLIPLoader/en.md (+3 -2)
📝 comfyui_embedded_docs/docs/CLIPLoader/es.md (+3 -2)
📝 comfyui_embedded_docs/docs/CLIPLoader/fa.md (+3 -2)

...and 80 more files

📄 Description

Summary

Daily node docs update.

New nodes (9, all with 14 translations):

MireloVideoToSFX / MireloTextToSFX — Mirelo SFX partner nodes (ComfyUI #17265)
VeedFabricNode, VeedLipsyncNode, VeedRemoveVideoBackgroundNode, VeedCleanAudioNode, VeedCleanVideoAudioNode — VEED partner nodes (ComfyUI #17242)
PadImageNode — mirror/color padding (ComfyUI #17224)
TextEncodeFibo2 — Fibo 2.0 text encoder (ComfyUI #17240)

Existing nodes whose source changed in a way the docs must follow:

Krea3ImageNode — new "ultra" thinking mode (several candidates, needs ≥1 image, always 2K, no precise_edit); the "auto" aspect ratio tooltip changed to "Krea picks the shape from the connected images and the prompt"
CLIPLoader — new fibo2 encoder type (qwen3-vl 4B) in the type list and the type-to-encoder table
BriaExpandImage — the expand logic now rejects an image that already covers the whole canvas and caps a ratio-expanded canvas at 25 megapixels
HeyGenVoiceTextToSpeechNode — the source no longer claims the turbo model costs half as much (the price badge changed), so the docs now say "costs less"

Script change: docs-generation/scripts/save_agent_doc.py gains COLOR in VALID_TYPES (IO.Color.Input, used by the Pad Image fill color).

Deliberately skipped

The scan reported 77 changed_nodes. Grouping by the last commit that touched each file and then running the black-list diff review left these as noise: chores: Update nodes names and categories (#16274), Fast disk detection (#16425), Remove torchaudio (#16457), [Partner Nodes] feat(client): … Idempotency-Key (#16220), price-badge rewrites, search_aliases, UI preview payloads (UI.SavedResult, background_images), internal renames (video_to_images_and_mask, qem_simplify), and the TextGenerate reasoning parser fix. None of them change an input, an output or a documented constraint, so nothing was rewritten for them.

Verification

Written by hand from the ComfyUI sources (comfy_api_nodes/nodes_mirelo.py, nodes_veed.py, nodes_krea.py, nodes_bria.py, nodes_heygen.py; comfy_extras/nodes_images.py, nodes_fibo2.py; nodes.py). English bodies validated with save_agent_doc.py (table shape, schema type names, fingerprint footer) and translated with batch_translate_docs.py for all 14 languages.

Local gates, run on the committed tree:

check_doc_structure.py --changed-only: 0 defects
check_md_format.py --changed-only: 180 files, 0 errors, 0 warnings
check_md_links.py --changed-only: all local resource links valid
fix_translation_preamble.py --check-only --changed-only: 0 leaked lines

Post-translation QC: fix_range_residue.py for all 14 languages (Range cells reverted to English by the translator: zh/zh-TW/ja/ko/ar/tr/pt-BR/fa/he/it/fr), unit words localized by hand (characters/seconds/images → per-language forms sampled from the existing corpus), check_translation_language.py 0 problems, and check_new_node_translations.py clean apart from three Italian overviews whose text is ASCII-only Italian (false positives of the ASCII check). update_param_translations.py was not run: the frontend has no node_translations.json entries for these node IDs (checked with sync_frontend_translations.py --export), so it would only have rewritten ~1700 unrelated files.

At 196 files this PR is over the 100-file review limit, so CodeRabbit will likely skip it; the local gates above stand in for that review.

No merge — for review only.

Follow-up (2026-10-23)

Added on top of the branch, in two commits:

docs: add TextEncodeMOSSImageEdit ... — new node from ComfyUI #17296 (Support MOSS-Image edit LoRAs): Text Encode MOSS-Image Edit with edit_format ("ai-toolkit", "ai-toolkit (kv_cache off)", "Moss-Image-Edit"), the growable images slot, and a note on which formats show the references to the text encoder (each reference downscaled to 384x384 with a Picture N: marker) versus latent-only.
docs: ... follow the MOSS edit, Fibo 2.0 negative prompt and MTP changes + docs: add 14-language translations ... — three schema changes merged after this PR was opened:
- FluxKontextMultiReferenceLatentMethod: reference_latents_method gained index_timestep_zero_isolated and index_timestep_zero_text_start (#17296). Range cell only, identifiers stay English in every language.
- TextGenerate: mtp gained the tree modes tree2s1, tree2s2, tree3s1, tree4s1m8, and the tooltip now describes breadth as well as depth (#17273). Per-language wording follows each file's existing sentence instead of a full retranslation.
- TextEncodeFibo2: negative_prompt tooltip now reads "for CFG with edit workflows", and a non-empty negative prompt is wrapped in Fibo 2.0's chat template instead of being cleaned like a caption (#17285). Row and fingerprint only.

Checks re-run on the committed tree, with the 225 files now in the diff:

fix_translation_preamble.py --check-only --changed-only: 0 files, 0 leaked lines
check_doc_structure.py --changed-only: 0 defects
check_md_format.py --changed-only: 225 files, 0 errors, 0 warnings
check_md_links.py --changed-only: all local resource links valid
check_new_node_translations.py / check_translation_language.py on TextEncodeMOSSImageEdit: clean, 0 problems
Table row counts equal to en.md in all 15 files of each touched node; one fingerprint line and one --- separator per file

update_param_translations.py was again not run: TextEncodeMOSSImageEdit has no entry in the exported frontend node_translations.json (checked with sync_frontend_translations.py --export), so it would only rewrite unrelated files; the node title comes from the source display_name.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/Comfy-Org/embedded-docs/pull/196 **Author:** [@lin-bot23](https://github.com/lin-bot23) **Created:** 10/21/2026 **Status:** ✅ Merged **Merged:** 10/24/2026 **Merged by:** [@comfyui-wiki](https://github.com/comfyui-wiki) **Base:** `main` ← **Head:** `daily-docs-update-1022` --- ### 📝 Commits (8) - [`8f68cf0`](https://github.com/Comfy-Org/embedded-docs/commit/8f68cf0a690f1d5839e9fe98b345054cd2be2793) Add docs for Mirelo, VEED, Pad Image and Fibo 2.0 nodes - [`64390f5`](https://github.com/Comfy-Org/embedded-docs/commit/64390f5738146a61e3cae51f4758168503cfdb02) docs: follow the Fibo 2.0 type, Krea ultra mode, Bria expand limits and HeyGen Voice pricing - [`36af12b`](https://github.com/Comfy-Org/embedded-docs/commit/36af12b81ba09b9696d8b88e0ce266ab9a4e6ee1) docs: add 14-language translations for the new nodes - [`f8bcdf8`](https://github.com/Comfy-Org/embedded-docs/commit/f8bcdf8a14a438b88578ee5323ab297f2bdc15b1) docs: add the fibo2 encoder type to CLIP Loader - [`dbe3567`](https://github.com/Comfy-Org/embedded-docs/commit/dbe356752be23079268703adfbe3a3ff81e56f5c) docs: add TextEncodeMOSSImageEdit and follow the MOSS edit, Fibo 2.0 negative prompt and MTP changes - [`0f695ab`](https://github.com/Comfy-Org/embedded-docs/commit/0f695aba4dda6c05f0a85656e9b5b5e40168aed9) docs: add 14-language translations for TextEncodeMOSSImageEdit and the changed rows - [`bd357ff`](https://github.com/Comfy-Org/embedded-docs/commit/bd357ff990f5f2cc84c5d9a207995ffcc9bb372a) Merge origin/main into daily-docs-update-1022 - [`329e0ed`](https://github.com/Comfy-Org/embedded-docs/commit/329e0edb0b8d2c1566f6dd6f2f9fd70b30e5dd09) Merge origin/main; update TextEncodeMOSSImageEdit for ComfyUI #17320 ### 📊 Changes **241 files changed** (+4309 additions, -211 deletions) <details> <summary>View changed files</summary> 📝 `comfyui_embedded_docs/docs/BriaExpandImage/ar.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/de.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/en.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/es.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/fa.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/fr.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/he.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/it.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/ja.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/ko.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/pt-BR.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/ru.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/tr.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/zh-TW.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/BriaExpandImage/zh.md` (+3 -3) 📝 `comfyui_embedded_docs/docs/CLIPLoader/ar.md` (+3 -2) 📝 `comfyui_embedded_docs/docs/CLIPLoader/de.md` (+3 -2) 📝 `comfyui_embedded_docs/docs/CLIPLoader/en.md` (+3 -2) 📝 `comfyui_embedded_docs/docs/CLIPLoader/es.md` (+3 -2) 📝 `comfyui_embedded_docs/docs/CLIPLoader/fa.md` (+3 -2) _...and 80 more files_ </details> ### 📄 Description ## Summary Daily node docs update. **New nodes (9, all with 14 translations):** - `MireloVideoToSFX` / `MireloTextToSFX` — Mirelo SFX partner nodes (ComfyUI #17265) - `VeedFabricNode`, `VeedLipsyncNode`, `VeedRemoveVideoBackgroundNode`, `VeedCleanAudioNode`, `VeedCleanVideoAudioNode` — VEED partner nodes (ComfyUI #17242) - `PadImageNode` — mirror/color padding (ComfyUI #17224) - `TextEncodeFibo2` — Fibo 2.0 text encoder (ComfyUI #17240) **Existing nodes whose source changed in a way the docs must follow:** - `Krea3ImageNode` — new `"ultra"` thinking mode (several candidates, needs ≥1 image, always 2K, no `precise_edit`); the `"auto"` aspect ratio tooltip changed to "Krea picks the shape from the connected images and the prompt" - `CLIPLoader` — new `fibo2` encoder type (`qwen3-vl 4B`) in the `type` list and the type-to-encoder table - `BriaExpandImage` — the expand logic now rejects an image that already covers the whole canvas and caps a ratio-expanded canvas at 25 megapixels - `HeyGenVoiceTextToSpeechNode` — the source no longer claims the turbo model costs half as much (the price badge changed), so the docs now say "costs less" Script change: `docs-generation/scripts/save_agent_doc.py` gains `COLOR` in `VALID_TYPES` (`IO.Color.Input`, used by the Pad Image fill color). ## Deliberately skipped The scan reported 77 `changed_nodes`. Grouping by the last commit that touched each file and then running the black-list diff review left these as noise: `chores: Update nodes names and categories` (#16274), `Fast disk detection` (#16425), `Remove torchaudio` (#16457), `[Partner Nodes] feat(client): … Idempotency-Key` (#16220), price-badge rewrites, `search_aliases`, UI preview payloads (`UI.SavedResult`, `background_images`), internal renames (`video_to_images_and_mask`, `qem_simplify`), and the `TextGenerate` reasoning parser fix. None of them change an input, an output or a documented constraint, so nothing was rewritten for them. ## Verification Written by hand from the ComfyUI sources (`comfy_api_nodes/nodes_mirelo.py`, `nodes_veed.py`, `nodes_krea.py`, `nodes_bria.py`, `nodes_heygen.py`; `comfy_extras/nodes_images.py`, `nodes_fibo2.py`; `nodes.py`). English bodies validated with `save_agent_doc.py` (table shape, schema type names, fingerprint footer) and translated with `batch_translate_docs.py` for all 14 languages. Local gates, run on the committed tree: - `check_doc_structure.py --changed-only`: 0 defects - `check_md_format.py --changed-only`: 180 files, 0 errors, 0 warnings - `check_md_links.py --changed-only`: all local resource links valid - `fix_translation_preamble.py --check-only --changed-only`: 0 leaked lines Post-translation QC: `fix_range_residue.py` for all 14 languages (Range cells reverted to English by the translator: zh/zh-TW/ja/ko/ar/tr/pt-BR/fa/he/it/fr), unit words localized by hand (`characters`/`seconds`/`images` → per-language forms sampled from the existing corpus), `check_translation_language.py` 0 problems, and `check_new_node_translations.py` clean apart from three Italian overviews whose text is ASCII-only Italian (false positives of the ASCII check). `update_param_translations.py` was not run: the frontend has no `node_translations.json` entries for these node IDs (checked with `sync_frontend_translations.py --export`), so it would only have rewritten ~1700 unrelated files. At 196 files this PR is over the 100-file review limit, so CodeRabbit will likely skip it; the local gates above stand in for that review. No merge — for review only. ## Follow-up (2026-10-23) Added on top of the branch, in two commits: - `docs: add TextEncodeMOSSImageEdit ...` — new node from ComfyUI #17296 (Support MOSS-Image edit LoRAs): `Text Encode MOSS-Image Edit` with `edit_format` (`"ai-toolkit"`, `"ai-toolkit (kv_cache off)"`, `"Moss-Image-Edit"`), the growable `images` slot, and a note on which formats show the references to the text encoder (each reference downscaled to 384x384 with a `Picture N:` marker) versus latent-only. - `docs: ... follow the MOSS edit, Fibo 2.0 negative prompt and MTP changes` + `docs: add 14-language translations ...` — three schema changes merged after this PR was opened: - `FluxKontextMultiReferenceLatentMethod`: `reference_latents_method` gained `index_timestep_zero_isolated` and `index_timestep_zero_text_start` (#17296). Range cell only, identifiers stay English in every language. - `TextGenerate`: `mtp` gained the tree modes `tree2s1`, `tree2s2`, `tree3s1`, `tree4s1m8`, and the tooltip now describes breadth as well as depth (#17273). Per-language wording follows each file's existing sentence instead of a full retranslation. - `TextEncodeFibo2`: `negative_prompt` tooltip now reads "for CFG with edit workflows", and a non-empty negative prompt is wrapped in Fibo 2.0's chat template instead of being cleaned like a caption (#17285). Row and fingerprint only. Checks re-run on the committed tree, with the 225 files now in the diff: - `fix_translation_preamble.py --check-only --changed-only`: 0 files, 0 leaked lines - `check_doc_structure.py --changed-only`: 0 defects - `check_md_format.py --changed-only`: 225 files, 0 errors, 0 warnings - `check_md_links.py --changed-only`: all local resource links valid - `check_new_node_translations.py` / `check_translation_language.py` on `TextEncodeMOSSImageEdit`: clean, 0 problems - Table row counts equal to `en.md` in all 15 files of each touched node; one fingerprint line and one `---` separator per file `update_param_translations.py` was again not run: `TextEncodeMOSSImageEdit` has no entry in the exported frontend `node_translations.json` (checked with `sync_frontend_translations.py --export`), so it would only rewrite unrelated files; the node title comes from the source `display_name`. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 22:10:34 +00:00