| English quick start | Full Chinese manual |
Choose Settings → Language / 语言 → English in the toolbox. The choice is saved and the page reloads. Version 2.36.0 extends the setting to workflow-node labels and major dialogs. Save the workflow, then refresh the ComfyUI page. Saved user content and external service responses are kept as written. Coverage and compatibility.
Click the gallery’s expand button for a draggable, resizable, minimizable workspace. Returning to the node keeps search, scroll and selection order. The single window shows its bound node; deleting that node or switching workflows pauses the binding. Images use their source aspect ratios in masonry columns. Counts sit beside tag names, while translated names and series appear below. Viewing a larger image does not select it.
Prompt Agent now shows raw WD14 tags or source descriptions directly above Final Prompt, expanded by default and collapsible. This read-only observation does not replace downstream output. Image instructions no longer have a 4000-character input cap. Output length is a configurable soft target (default 4000; 0 disables it); oversized results remain complete, including JSON. Model context and token limits still apply.
| Tool | Main uses |
|---|---|
| Batch LoRA loader | Parse <lora:name:weight> syntax, visually choose local models, adjust weights, enable/reorder entries, reuse groups and output activated trigger words |
| Local LoRA | Recursive scans with relative paths, subfolder categories, Civitai matching, previews, metadata, editable trigger words and sending models to ComfyUI |
| LoRA Explorer | Browse, filter, discover and download Civitai models |
| Prompt Library / Cards | Save, search, categorize and edit prompts; bilingual tag cards, hide/restore fragments, translation, normalization and Chinese/English autocomplete |
| Prompt Agent | Anima / Krea 2 prompt creation and iterative edits, locks, undo/history and image interrogation/expansion through configured services |
| Prompt Batch | Sequential groups from input/prompts/, with server-side progress, pause/resume, retry, skip and cancel |
| Other nodes | Camera/pose references, outfit selection, prompt saving, text joining, string/image routing and Anima/Cosmos empty latent presets |
Prompt Agent requires your own compatible model API. Optional DSH integration is separate from normal API use. Models and personal credentials are not included.
Multi Gallery supports Danbooru, Civitai, Pixiv, Safebooru, yande.re and Konachan.net, with source-specific search, filters, previews, downloads and image/metadata output. Authorization requirements differ by source. Pixiv reference images do not provide generation prompts. The separate local AnimaDex picker inserts character prompt tags without an LLM.
Settings includes language, appearance, typography, custom CSS and data-transfer controls. The toolbox menu reopens release notes and the bridge download.
| Data | Location |
|---|---|
| Language, appearance and browser preferences | Current browser profile and site storage |
| Outfit cards | Browser clothing-db IndexedDB; use the library’s own export/import |
| Prompt Library | IndexedDB plus data/prompt_library.json and its backup |
| LoRA groups/categories and shared metadata | data/anima_meta.json |
| Source credentials/provider configuration | Local files under data/; not shipped with the plugin |
| Prompt batch manifests | data/batches/ |
Use a consistent browser profile and address/port. localhost, 127.0.0.1 and different ports use separate browser storage. Export the relevant settings/libraries before clearing site data, and keep data/ when updating or moving an installation. Directory permissions may need to be granted again.
The interface language switch currently covers the toolbox only. Workflow nodes and remaining plugin interfaces are deferred to a later release. Prompt Agent translation checks are advisory: a valid layout can continue with a visible quality warning, and translation completion is not guaranteed. Layered region painting and mask-to-bbox editing are not included in this release.