VideoFreeStudio

# VideoFree Studio **English** · [中文](README_ZH.md) Current version: `2.0.0` Official website: [videofreestudio.com](https://videofreestudio.com) **Turn fragmented local AI workflows into complete creative capabilities that people can use, developers can distribute, and agents can call reliably.** VideoFree Studio is a local AI video creation workspace, capability gateway, and plugin platform. It uses ComfyUI as a generation engine without forcing users to work directly with complex node graphs, model paths, and disconnected workflow files. Models, workflows, parameter forms, prompt expertise, and supporting assets are packaged as installable plugins, while projects, categories, assets, and jobs are managed in one place. People can create in the workspace, while external agents use the same plugin capabilities, project data, and local media through MCP or HTTP APIs. ## Why VideoFree Studio ### Local AI is powerful, but fragmented Cloud products choose the model, parameters, and interaction flow for the user. Moving to local creation means facing an overwhelming number of models, workflows scattered across the web, hard-to-read ComfyUI graphs, missing custom nodes, incompatible versions, and model paths that differ from one installation to another. VideoFree Studio does not replace ComfyUI. It adds a stable product layer on top of it. A tested workflow can be packaged with its input rules, parameters, compiler logic, UI, dependencies, and documentation. Users select a capability like they would in a regular creative application instead of learning its internal graph first. ### Every model speaks a different prompt language Image, video, speech, and music models use different prompt formats, media roles, and parameter constraints. The same request may need a completely different execution prompt for each model. A VideoFree Studio plugin can ship model-specific collaboration skills. Users describe the desired scene, motion, dialogue, or music in plain language; the collaboration flow asks for missing information, checks media relationships, and compiles the request into the format expected by the selected model. Model-specific knowledge stays inside the plugin instead of accumulating in the framework. ### Local files, prompts, and production notes quickly become unmanageable ComfyUI executes workflows well, but it is not a complete video project manager. Images, clips, voices, music, intermediate files, scripts, prompts, references, and chat notes often end up in unrelated folders. VideoFree Studio organizes creation as **project → category → asset**. A project can represent a film or series, while categories can represent episodes, chapters, or shot groups. Final media is archived automatically. Job records preserve source media, parameters, capability identity, and regeneration lineage. Reusable characters, scenes, and voices can be maintained as project assets. Project plugins can also manage scripts, storyboards, and progress in the current project or category context. ### Agents need a stable capability contract, not raw graph access Directly manipulating ComfyUI graphs is brittle for agents: node IDs, connections, model filenames, and workflow versions can change at any time. VideoFree Studio exposes workflows as capabilities with stable IDs, input schemas, and execution contracts. An agent discovers capabilities, uploads media, submits validated inputs, and monitors jobs. VideoFree Studio handles workflow selection, media binding, queues, recovery, error reporting, and result archiving. The workspace, MCP, and HTTP API all share one capability catalog and one job service. ### Workflows become installable products A single plugin can contain: - ComfyUI workflows; - model and capability declarations; - parameter forms; - workflow compilers and native executors; - model-specific prompt and collaboration skills; - presets, icons, voice samples, and other fixed assets; - post-processing actions; - project tools or a complete application UI. The same plugin can serve people in the workspace and agents through MCP or HTTP. Plugin development resources: - [Plugin development guide](plugins/PLUGIN_DEVELOPMENT.md) - [Plugin protocol](docs/plugin-protocol.md) - [VideoFree Studio plugin development skill](.agents/skills/videofree-plugin-development/SKILL.md) ## Core capabilities - Local and cloud image, video, speech, and music generation. - Declarative capability discovery and parameter validation. - Model-specific prompt compilation and collaboration. - Persistent queues, cancellation, restart recovery, and job lineage. - Project/category media library and reusable project assets. - Resource actions such as refinement, upscaling, interpolation, and image splitting. - Application plugins and contextual project tools. - Shared access through the workspace, MCP, and HTTP API. ## Quick start VideoFree Studio supports Windows release packages and Linux source deployments. 1. Prepare and start ComfyUI with the models and custom nodes required by the plugins you want to use. 2. Extract the complete VideoFree Studio release package. 3. On Windows, double-click `videofree.bat`. On Linux, install `requirements.txt`, run `chmod +x videofree.sh`, then start with `./videofree.sh`. 4. Open the workspace, create a project, and enter the actual ComfyUI URL in Settings. The common default is `http://127.0.0.1:8189`, but your installation may use another port. 5. Start with a low-resolution image or short clip to verify the environment before running expensive jobs. The release package includes its Python runtime and does not require a separate Python or Node.js installation. Source deployments must install `requirements.txt` and build the frontend. The editing plugin requires `ffmpeg` and `ffprobe` on system `PATH`; these binaries are not stored in this repository. Download FFmpeg from [the official download page](https://ffmpeg.org/download.html), then verify both `ffmpeg -version` and `ffprobe -version` in the environment that launches VideoFree Studio. See the [editor installation guide](plugins/film-studio/INSTALL.md) for platform-specific instructions. The optional companion ComfyUI package is currently hosted on Quark Drive. Copy the complete share text into the Quark app if the browser link does not open: - Share text: `/~2ca93bYCBE~:/` - URL: <https://pan.quark.cn/s/2d145a999169> Keep model filenames and subdirectories unchanged. Some bundled workflows require SageAttention installed in the Python environment actually used by ComfyUI. Do not install it only into the system Python or upgrade PyTorch/CUDA without checking compatibility. ## Configuration Both `videofree.bat` and `videofree.sh` read a root `.env` file. The main options are: ```dotenv COMFYUI_URL=http://127.0.0.1:8189 MEDIA_GATEWAY_HOST=0.0.0.0 MEDIA_GATEWAY_PORT=8765 MEDIA_GATEWAY_TOKEN= ``` The access token is empty by default. The token saved in Settings is shared by the workspace, HTTP API, and MCP and takes precedence over the environment default. Do not expose an unauthenticated gateway directly to the public internet. Project media is stored under `data/projects/` unless another library directory is selected in Settings. System settings, projects, and job records are stored in `data/media_gateway.db`. Plugin-owned secrets are stored in each plugin's ignored `.env` file. Fixed plugin resources must remain inside the plugin directory. ## Agent and MCP access The MCP endpoint starts with the gateway: ```json { "mcpServers": { "videofree": { "type": "http", "url": "http://127.0.0.1:8765/mcp" } } } ``` For another machine on the LAN, replace `127.0.0.1` with the gateway machine's real LAN address. If a gateway token is configured, add it as the `X-API-Token` header. Client naming for HTTP MCP transports differs; use the transport name required by your client. Agents should begin with health and discovery calls, then submit only capability IDs and schemas returned by the running gateway. They should not infer model paths, workflow files, or ComfyUI nodes. - MCP guide: [examples/skills/videofree-creation/references/mcp-install.md](examples/skills/videofree-creation/references/mcp-install.md) - Direct API skill: [examples/skills/videofree-creation](examples/skills/videofree-creation) - Interactive API documentation: `http://127.0.0.1:8765/docs` - Capability catalog: `http://127.0.0.1:8765/api/v2/capabilities` ## Plugin installation and development Each plugin is a complete `plugins/<plugin-id>/` folder. Its manifest, workflows, Python handlers, collaboration skills, UI, icons, presets, and fixed assets must stay inside that folder. Wait for jobs using the plugin to finish, copy or update the whole folder, restart the gateway, and refresh the workspace. Python plugins are trusted local code, not a sandbox. Plugins do not automatically install models, ComfyUI nodes, or Python packages. Use the included skill when asking a coding agent to create, migrate, review, or validate a plugin: > Use `$videofree-plugin-development` to package this ComfyUI workflow as a VideoFree Studio plugin. Keep all model-specific code and assets inside the plugin, expose a stable capability schema, and run the plugin and release checks. ## Community - AI Arsenal and project updates: [laobanclaw.top](https://laobanclaw.top/) - **Knowledge Planet**: AI video automation workflows and practical knowledge about AI, agents, and large language models. - **VideoFree Studio WeChat group**: discuss local models, plugin development, workflow packaging, and AI video creation. If the QR code expires, visit [laobanclaw.top](https://laobanclaw.top/) for the latest entry. <table> <tr> <td align="center"><strong>Knowledge Planet</strong><br /><a href="frontend/public/contact/zsxq-qr.png"><img src="frontend/public/contact/zsxq-qr.png" alt="Knowledge Planet QR code" width="220" /></a></td> <td align="center"><strong>VideoFree Studio WeChat group</strong><br /><a href="https://laobanclaw.top/videofree/qrcode.jpg"><img src="https://laobanclaw.top/videofree/qrcode.jpg" alt="VideoFree Studio WeChat group QR code" width="220" /></a></td> </tr> </table> ## License and third-party boundaries Project attribution, release evidence, commercial delivery records, and contributor authorization practices are documented in [Source provenance](docs/source-provenance.md). See [NOTICE](NOTICE) for project source and copyright notices. VideoFree Studio-owned source code is released under [GNU AGPL-3.0](LICENSE) ([Chinese explanation](LICENSE_ZH.md)). For closed-source integration, enterprise deployment, white-label use, support, or custom development, see the [commercial licensing notice](COMMERCIAL_LICENSE.md). ComfyUI, community nodes, models, LoRAs, fonts, FFmpeg, and other third-party resources keep their own licenses. The editor is the `plugins/film-studio/` application plugin; it requires user-installed `ffmpeg` and `ffprobe` on system `PATH` and does not redistribute those binaries. See its [installation guide](plugins/film-studio/INSTALL.md). `examples/skills/openmontage/` is an independent automation skill and is not a runtime dependency of the editor. See [third-party notices](THIRD_PARTY_NOTICES.md). Only install plugins from sources you trust. Model and sample-file redistribution rights must be verified separately from VideoFree Studio's source-code license.