(official name coming soon™ — we're open to suggestions)
A local-first, AI-native video editor for generative content creation.
Bring Your Own Workflow. Keep your data. Own your creative pipeline.
What's This? • Status • ComfyUI • Under the Hood • Get Involved
NLA AI Video Creator is an open-source desktop app that bridges the gap between AI generation tools and video editing. If you've ever found yourself:
- Juggling between ComfyUI, file explorers, and video editors
- Manually renaming and organizing generated assets
- Wishing you could see your AI-generated clips on a timeline with your audio
- Wanting to iterate on generations without losing your creative flow
...then this project is for you.
A purpose-built timeline editor where:
- 🎵 Audio, images, and video live together — See your soundtrack alongside AI-generated visuals
- 🔌 ComfyUI is a first-class citizen — Connect your local workflows directly to the editor
- 🧠 Generation happens in-context — Select a clip, tweak parameters, hit generate, see results
- 💾 Everything stays local — Your projects, your machine, your data
Philosophy: This isn't trying to replace Premiere or DaVinci. It's the missing link between "I have cool AI workflows" and "I have a finished video."
This is a passion project in early stages. Things work, things break, APIs change. If you're looking for a polished tool to use today, check back later!
If you're here to:
- ⭐ Watch the project evolve
- 🛠️ Contribute code or ideas
- 🧪 Experiment with early builds
...you're in the right place. Star the repo to follow along!
| Feature | Status |
|---|---|
| Timeline with tracks (video/audio/markers) | ✅ |
| Drag, resize, and manage clips | ✅ |
| GPU-accelerated preview with transforms | ✅ |
| ComfyUI workflow integration (image gen) | ✅ |
| Generative assets with version history | ✅ |
| Provider Builder UI (no JSON editing required) | ✅ |
| Project save/load | ✅ |
- Audio playback & waveform visualization
- Video generation workflow support (backend mostly complete)
- Smart input suggestions (timeline as implicit wiring)
- More provider adapters (fal.ai, Replicate, etc.)
- Export to video file
- macOS & Linux builds
See the full roadmap in docs/PROJECT.md.
This is where things get interesting. Bring Your Own Workflow™ — your ComfyUI setups become first-class providers in the editor.
- Point the app at your local ComfyUI instance
- Use the Provider Builder to select which workflow inputs to expose (prompts, seeds, steps, CFG, etc.)
- Bind parameters via a visual node browser — no JSON editing required
- Generate directly from the timeline — results land in your project with version history
No vendor lock-in. No cloud dependency. Your workflows, your way.
ComfyUI has become the power-user's playground for AI image and video generation. But it's a workflow tool, not an editing tool. This project aims to be the bridge — letting you orchestrate your ComfyUI outputs in a timeline-based environment without leaving your creative flow.
The provider system is designed to be extensible. ComfyUI is the first adapter, but the architecture supports:
- Custom HTTP endpoints
- Commercial APIs (fal.ai, Replicate, etc.) — planned
- Any backend that can accept parameters and return media
For the developers curious about what makes this tick — the preview and compositing pipeline is where we've invested significant effort. Here's the architecture:
The challenge: Dioxus runs in a WebView (WebView2 on Windows), but we need GPU-accelerated video compositing. Our solution bypasses WebView limitations entirely.
┌─────────────────────────────────────────────────────────────────┐
│ Preview Pipeline │
├─────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │ FFmpeg │ │ Frame │ │ wgpu Native │ │
│ │ Decode │── ▶│ Cache │──▶│ Compositor │ │
│ │ Workers │ │ (LRU) │ │ Surface │ │
│ └──────────────┘ └──────────────┘ └──────────────────┘ │
│ │ │ │ │
│ ▼ ▼ ▼ │
│ • In-process decode • 8GB budget • Layer stacking │
│ • HW accel (D3D11VA) • Prefetch window • Per-clip xforms │
│ • Parallel workers • Latest-wins • GPU compositing │
│ • CPU fallback • Per-asset keying • Native surface │
│ │
└─────────────────────────────────────────────────────────────────┘
Three key components:
-
FFmpeg Decode Workers — In-process video decoding via
ffmpeg-next. Supports hardware acceleration on Windows (D3D11VA/DXVA2) with automatic CPU fallback. Parallel decode workers keyed by track to avoid decoder contention. -
Frame Cache — LRU cache with an 8GB budget for smooth scrubbing. Prefetch windows (5s ahead, 1s behind) warm the cache when idle. Latest-wins scheduling cancels stale decode jobs when you scrub quickly — only the frames you need get decoded.
-
wgpu Native Compositor — A native GPU surface that renders alongside the WebView. Each layer uploads as a texture; shaders handle transforms (position, scale, rotation, opacity) and compositing. This bypasses the DOM entirely for zero-copy GPU rendering.
We built in visibility for optimization work:
- Preview stats overlay with per-stage timing (seek, decode, transfer, scale, upload)
- Hardware decode percentage — see how much is offloaded to the GPU
- Cache hit rate tracking — know when you're hitting vs. missing
- HW Dec toggle — force CPU decode for A/B comparisons
Work in Progress: We're still optimizing. The GPU currently receives RGBA after CPU conversion — a future path keeps YUV/NV12 on the GPU to avoid the round-trip. There's headroom to improve.
| Component | Technology |
|---|---|
| Language | Rust — Fast, safe, no runtime |
| UI Framework | Dioxus 0.7 — Reactive, cross-platform, hot-patching |
| GPU Rendering | wgpu — WebGPU-based, cross-platform compositing |
| Video Decode | FFmpeg (ffmpeg-next) — In-process decode with HW accel |
| Async | Tokio — Background tasks, provider communication |
Detailed docs live in the /docs folder:
- PROJECT.md — Vision, architecture, roadmap, and session changelog
- CONTENT_ARCHITECTURE.md — How assets, generation, and the timeline work together
- PROVIDER_SETUP_GUIDE.md — Setting up ComfyUI and other providers
- DECODE-STRATEGIES.md — Deep dive on NLE preview pipeline architecture
📝 Full setup guides coming soon. For now, adventurous developers can explore the docs and source code.
This is an open source project and contributions are welcome!
Ways to help:
- ⭐ Star the repo — Helps visibility
- 🐛 Report issues — Found a bug? Let us know
- 💡 Suggest features — Open a discussion
- 🔧 Contribute code — PRs welcome
- Provider adapters for other services (fal.ai, Replicate, etc.)
- macOS and Linux testing/builds
MIT License — See LICENSE for details.
Use it, fork it, build on it. 🤖
Built with 🦀 Rust and ☕ too much coffee