# comfyui-wenwu **Repository Path**: ai-tools_4/comfyui-wenwu ## Basic Information - **Project Name**: comfyui-wenwu - **Description**: No description available - **Primary Language**: Unknown - **License**: Not specified - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2026-07-18 - **Last Updated**: 2026-07-18 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # comfyui-wenwu ComfyUI custom nodes for WenWu local prompt workflows, covering Bernini image/video prompts, an independent LTX-2.3 multi-reference prompt enhancer, Krea2 prompt rewriting, and ACE Step 1.5 music planning. This node package is built for local generation inside ComfyUI. It lets users enter natural spoken-style descriptions and then uses local `llama.cpp` inference through `ComfyUI-llama-cpp` to convert them into structured prompts for different downstream tools. The goal is to save users from manually writing strict prompt text while keeping the conversion local and reusable across visual and music workflows. ## What this node provides - **Bernini Prompt wenwu**: converts natural user descriptions into professional Bernini-oriented prompt text for image and video tasks through a local `ComfyUI-llama-cpp` model. - **WenWu LTX-2.3 Prompt Enhancer**: generates duration-aware English LTX-2.3 video prompts from natural-language direction and as many as five reference images. References 1-4 are subject slots and Reference 5 is the dedicated Licon background slot. - **WenWu Krea2 Prompt Instruct**: converts Chinese or mixed-language image ideas into a single polished English Krea2 prompt. It supports text-to-image prompt expansion and optional style-reference prompt rewriting. - **ACE Step 1.5 LLM WENWU**: converts natural music descriptions into ACE Step 1.5-ready tags, lyrics, BPM, duration, language, key, and conditioning. It includes a preview/refresh UI and an editable confirmed prompt box. - **Simple Text**: a lightweight canvas text node that can be edited by double-clicking the node or using the node context menu. ## Model-focused workflows This package is not a single Bernini node. It contains four model-facing prompt helpers that share the same WenWu idea: write naturally first, then let a local LLM turn the request into the stricter format expected by the target model. | Node | Target model or workflow | Best for | Main output | | --- | --- | --- | --- | | **Bernini Prompt wenwu** | Bernini image and video prompt workflows | Turning loose Chinese or mixed-language scene ideas into polished Bernini-style visual prompts | Structured prompt text for image/video generation | | **WenWu LTX-2.3 Prompt Enhancer** | LTX-2.3 and Licon MSR multi-reference video workflows | Grounding subjects and a dedicated environment from up to five images while creating a duration-aware cinematic prompt | English LTX-2.3 prompt plus width, height, frame count, and frame rate | | **WenWu Krea2 Prompt Instruct** | Krea2 text-to-image and style-reference workflows | Rewriting short image ideas into a single continuous English Krea2 prompt, with optional style-reference image guidance | Clean English Krea2 prompt text | | **ACE Step 1.5 LLM WENWU** | ACE Step 1.5 music generation workflows | Planning songs from natural music descriptions, including style, lyrics, vocals, BPM, key, duration, and conditioning | ACE Step-ready tags, lyrics, prompt text, `conditioning`, and `duration` | ## Current status This repository is synced from a local ComfyUI custom node directory: ```text E:\ComfyUI-v3\ComfyUI\custom_nodes\comfyui-wenwu ``` Recent local changes replace the previous `WenWuShowAndSaveAnything` node export with `WenWuSimpleTextNode`. The old `WenWuShowAndSaveAnything` implementation is still present in `wenwu_prompt_generator.py`, but it is no longer registered in `nodes.py`. `Simple Text` is implemented mostly in `js/showAnything.js`. It stores node text and style values in node properties and supports: - multiline text - font size, color, weight, and italic style - left, center, and right alignment - optional background color - padding and border radius - color swatches in the editor `ACE Step 1.5 LLM WENWU` is implemented in `ace_step_llm_wenwu.py` with frontend controls in `js/ace_step_plan_preview.js`. It supports: - natural-language music planning through a connected llama.cpp model loader - preview/refresh prompt generation without running the full audio workflow - editable confirmed prompt mode - instrumental/no-vocal handling - automatic TAGS line formatting - style-based BPM, time signature, language, and key suggestions - direct `conditioning` and `duration` outputs for ACE Step 1.5 workflows `WenWu Krea2 Prompt Instruct` is implemented in `krea2_prompt_instruct.py` with frontend controls in `js/krea2_prompt_instruct.js`. It supports: - local Krea2 prompt generation through a connected llama.cpp model loader - Chinese or mixed-language input rewritten into one continuous English prompt - optional style-reference mode with an uploaded image used only for transferable visual style - automatic hiding of the style image selector when style-reference mode is off - clean output suitable for direct Krea2 text-to-image generation `WenWu LTX-2.3 Prompt Enhancer` is implemented independently in `ltx23_wenwu_node.py`, with its preview and edit controls in `js/ltx23_prompt_preview.js`. It supports: - one to five reference images in a single multimodal Llama pass - fixed Licon mapping: References 1-4 supply subjects and Reference 5 supplies the exclusive background environment - subject identity separation to reduce face, hairstyle, costume, accessory, voice, and role blending - exclusion of incidental portrait backgrounds unless the user explicitly asks to retain one - English cinematic narration while preserving quoted dialogue exactly in its original language and script - duration-aware story pacing, shot count, action, camera direction, synchronized dialogue, ambience, effects, and optional background music - editable enhanced-prompt preview: locked mode regenerates with Llama; edit mode uses the confirmed prompt without another Llama inference - configurable duration, frame rate, resolution, seed, and LTX-compatible frame-count output - automatic llama.cpp/mmproj memory cleanup before downstream video models run - vision-thumbnail acceleration applied only to the Llama analysis copy: subject references are limited to a 640-pixel longest edge and Reference 5 to 768 pixels; original images remain untouched for Licon MSR and downstream LTX-2.3 processing The thumbnail optimization fixes a previous single-image encoding path that accepted a size limit but did not actually resize separately connected reference images. It improves multi-reference prompt inference speed without lowering the resolution of images used by the video-generation workflow. Very small facial marks, jewelry engravings, or fine textile details may still benefit from clearer source images and explicit natural-language identity anchors. ## Version notes See [CHANGELOG.md](CHANGELOG.md) for the current update notes. The latest update clarifies the target workflows, documents the Krea2 and LTX-2.3 nodes separately from the Bernini node, and makes optional model-specific imports more tolerant so one missing workflow dependency does not prevent the rest of the node pack from loading. ## Installation Clone this repository into the ComfyUI `custom_nodes` directory: ```bash cd ComfyUI/custom_nodes git clone https://github.com/liaodaobin/comfyui-wenwu.git ``` Install Python dependencies: ```bash pip install -r comfyui-wenwu/requirements.txt ``` Restart ComfyUI after installation. ## Dependencies The prompt generator expects `ComfyUI-llama-cpp` to be installed next to this node directory: ```text ComfyUI/custom_nodes/ComfyUI-llama-cpp ComfyUI/custom_nodes/comfyui-wenwu ``` If you want this plugin to be copyable as a single folder, copy both directories together as siblings. Python dependencies are listed in `requirements.txt`. The Krea2 node depends on `ComfyUI-llama-cpp` for local prompt rewriting. The ACE Step music node also expects an existing ACE Step 1.5 ComfyUI setup, including the required ACE Step nodes and model files. This repository does not include ACE Step model weights, llama.cpp model weights, or ComfyUI itself. ## Notes - Runtime logs are written under `logs/` and are ignored by git. - Python cache files are ignored by git. - The current repository does not include model weights or ComfyUI itself.