Framework for Unified Kreation
Framework for Unified Kreation
Local-first rendering pipeline for professional AI generation
Get StartedClone the repo and run the setup script to get FUK running on your machine. The wiki walks through installation step by step.
View on GitHubA more user-friendly installer is coming soon.
FUK is in active development. Core functionality is working and the app is stable, but AOV outputs need refinement, and it needs beta testing.
Contributions, testers, and suggestions are highly appreciated.
Currently focusing on: Improving AOV quality, quality of life improvements, localized data collection, speed and optimization. Before moving on to big features planned for later.
Professional AI generation tools shouldn't force you to work around them. FUK integrates into your existing workflows with proper project organization, native lossless EXR support, and stable infrastructure that doesn't break with every update.
Your hardware. Your models. Your data. No mandatory cloud services. No compromises.
What you'll need to run FUK effectively on your local hardware.
Image and video generation with proper project/shot/version organization. Industry-standard hierarchies that integrate with existing pipelines.
Depth maps, normal passes, cryptomattes. Multi-layer 32-bit EXR export ready for professional applications.
Preprocessor integration for OpenPose, depth, edge detection and regional conditioning via masked zones. Generate with precision using external image inputs.
Upscaling via Real-ESRGAN or SeedVR2 and frame interpolation via FILM. Professional quality enhancement for generated content.
Runs entirely on your hardware. No telemetry, no cloud requirements, no surprise authentication. Your data stays yours.
Built on stable APIs without fragile dependencies. Tools that work today will work tomorrow without breaking your pipeline.
Recently landed features. Full details are in the wiki.
Plan a whole shot list before you generate a frame. Panels carry their own prompt, mood, and seed, with project-wide tags and active LoRAs shared across them.
Read more in the wiki →Type # in any prompt field to autocomplete reusable markers for characters, props, locations, and style — expanded automatically at generation time.
Point it at a folder of images, pick a content preset, and it curates and captions a training-ready dataset automatically.
Read more in the wiki →Drive FUK straight from Blender's viewport — scene-accurate control passes, live IPR-style preview, and VACE video generation from your camera moves.
Read more in the wiki →SDR → HDR via the LTX-2.5 IC-LoRA rebuilds clipped highlight range, and ACES color management via OpenColorIO writes ACEScg or ACES2065-1 for your facility's config. All of it lands in multi-layer 32-bit float EXR.
Read more in the wiki →Active development with a clear direction. No hype, just what's actually being worked on.