Install Plugins
Copy a folder, restart the gateway — the plugin is installed. You don't fetch models one by one either: hand the plugin folder to an AI agent and let it install everything per INSTALL.md. This guide covers installation, model download, and verification, with a quick-reference table for every demo plugin.
1Where plugins come from
- Bundled demo plugins: the image, video, speech, music, and post-production plugins under
plugins/are demos showing what one plugin can package. - Plugins shared by others: any complete plugin folder installs the same way.
- Your own plugin: see Build Plugins.
2Install: copy the folder, restart the gateway
VideoFree/
└─ plugins/
└─ your-plugin/ ← copy the whole folder in; plugin.json is the only registration entry
After restarting the gateway, the capability appears in the workspace UI and the MCP capability list — no integration code needed. Uninstall = remove the folder and restart.
Even this step is a one-liner for your agent: “Copy [plugin folder] into VideoFree’s plugins/ directory, then restart the gateway.”
3Models and nodes: let your agent install per INSTALL.md
Plugins never download models by themselves — and you don’t have to fetch them one by one either. Each plugin’s INSTALL.md declares the exact model filenames, official download URLs, and target directories. Hand the plugin folder to an agent that can operate your computer (Claude Code, Codex, OpenCode, …) and let it read the file and install everything.
Send this prompt, replacing the paths with your own:
I want to enable a VideoFree plugin: [plugin folder path, e.g. C:\VideoFree\plugins\ltx-2.5]. First read INSTALL.md in that folder and list every model file and ComfyUI node package it requires. My ComfyUI root is: [your ComfyUI path]. Check which files already exist (exact filename and subfolder), download the missing ones from the official URLs declared in INSTALL.md, and place them in the declared directories. Do not rename files or flatten directories. When everything is in place, tell me to restart ComfyUI and list what you downloaded and what you skipped. If a license gate or download failure blocks you, stop and explain.
The Chinese version of this prompt lives on the 插件安装 page.
4Verify the installation
- Restart ComfyUI — the startup log must show no
IMPORT FAILED. - Run ComfyUI Manager’s missing-node check against the node packages the plugin declares.
- Submit one minimal job in VideoFree (a low-resolution image or a 480P clip). Media out means the plugin is installed.
If you download models manually, three rules: filenames and subfolders must match INSTALL.md exactly (workflows load files by exact name; renaming or flattening causes “missing model” errors); download only from the declared official sources; restart ComfyUI, then verify. If VRAM is tight, prefer int8 / fp8 quantized files, lower resolution and duration first, and raise quality only after the pipeline works.
Appendix: what each plugin needs
The table below is a quick index; the plugin’s own INSTALL.md is always the authoritative list.
| Plugin | Type | Model sources | ComfyUI nodes |
|---|---|---|---|
| Krea2 (image) | Local | Comfy-Org/Krea-2: krea2_turbo_fp8_scaled, qwen3vl_4b_fp8_scaled, qwen_image_vae; identity views also need krea2_identity_edit_v1_2 (comfyui-krea2edit); upscale also needs OmniSR_X4_DIV2K | comfyui-krea2edit, ComfyUI-Easy-Use, rgthree-comfy, ComfyUI_essentials, ComfyUI_UltimateSDUpscale |
| Qwen Image 2.1 (image) | Local | QwenLM/Qwen-Image-2.1: qwen_image_2.1_int8_convrot (into models/diffusion_models/Qwen/), qwen3vl_8b_int8_convrot, qwen_image_2.1_vae_bf16 | Up-to-date ComfyUI core |
| GPT-Image (cloud image) | Cloud | No local models; configure a GrisAI API key in plugin settings | None |
| MiniMax H3 (video) | Local | MiniMax-AI/MiniMax-H3 official project: fl2va / ref2va backbones, qwen3vl_32b_minimax_h3_nvfp4_awq, video & audio VAEs, turbo LoRAs, latent upscaler; the LTX refine stage also needs the full LTX-2.5 set | Up-to-date ComfyUI core; LTX refine needs ComfyUI-LTXVideo |
| LTX 2.5 (video) | Local | Lightricks/LTX-2.5: int8 transformer, Gemma text encoder, video/audio VAEs, spatial upscaler, multi-reference LoRA | ComfyUI-LTXVideo, VideoHelperSuite, KJNodes |
| CosyVoice3 (speech) | Local | Fun-CosyVoice3-0.5B, auto-downloaded by the node on first use | ComfyUI_Fill-Nodes |
| Qwen3-TTS (speech) | Local | Qwen3-TTS 1.7B repos (VoiceDesign / Base / CustomVoice), auto-downloaded by Qwen3Loader | ComfyUI-Qwen3-TTS |
| ACE-Step 1.5 (music) | Local | ACE-Step/ACE-Step-v1.5: turbo, two text encoders, VAE | Up-to-date ComfyUI core |
| MiniMax Music 3 (music) | Local | MiniMax-AI/MiniMax-Music3 official project: DiT, text encoder, DAV | Up-to-date ComfyUI core |
| Video post-processing (upscale/interpolation) | Local | LTX-2.5 upscale set (into models/*/LTX-2.5/ subfolders), deblur LoRA, RIFE (auto-downloaded by the node) | ComfyUI-Frame-Interpolation, VideoHelperSuite, KJNodes, Custom-Scripts, Easy-Use, LTXVideo |
| Editor film-studio (final cut) | Local | Auto subtitles use whisper-large-v3-turbo (auto-downloaded by the node); system needs ffmpeg and ffprobe | ComfyUI-Whisper, rgthree-comfy |
| Image split / Project archive / AI progress | Tool | No models, no node dependencies | None |