Fizgig
Fizgig — LoRA & Fine-tune Studio for Klein 9B, Krea 2, MiniMax H3, Qwen Image 2.1, Anima & Z-Image Turbo and Ace-Step XL: train, fine-tune, profile, repair and
What is Fizgig?
Fizgig — LoRA & Fine-tune Studio for Klein 9B, Krea 2, MiniMax H3, Qwen Image 2.1, Anima & Z-Image Turbo and Ace-Step XL: train, fine-tune, profile, repair and extract LoRAs & LoKR for Image, Video & Music Models
Overview of Fizgig
What it is
Fizgig is a trainer and workbench for image, video, and music models. It runs in its own window on your computer, in your browser, or on a rented GPU, and you can monitor a run from your phone. Every trainer makes LoRAs; Fizgig also provides tools to fix, explore, compare, and profile LoRAs after training. It supports multiple models including Flux 2 Klein 9B, Flux 3 Image Klein, Krea 2, MiniMax H3, Qwen Image 2.1, MOSS-Image, Z-Image Turbo, SDXL, Anima, and ACE-Step. Memory management is automatic: precision, block swap, and previews adjust to available VRAM. Fizgig loads LoRAs from kohya, PEFT, OneTrainer, AI-Toolkit, and LyCORIS, and saves .safetensors files compatible with ComfyUI. It is free and open source under the Apache-2.0 license.
Key features
- Train LoRAs, LoKRs, slider LoRAs, edit LoRAs, and full fine-tunes on image, video, and music models.
- Repair a LoRA block by block without retraining: drag a slider per block and save a new .safetensors.
- LoRA Royale: render every epoch on one seed with a crossfade and export shareable loops to find the best epoch by eye.
- Profile which blocks carry identity, style, and detail before any modification.
- Fizgig LoRA Isolation: train two subjects into one LoRA while preserving each subject's identity.
- Memory auto-plans precision, block swap, and previews to fit your free VRAM; if a preview cannot fit, training continues.
- Supports models: Flux 2 Klein 9B, Flux 3 Image Klein, Krea 2, MiniMax H3, Qwen Image 2.1, MOSS-Image, Z-Image Turbo, SDXL, Anima, and ACE-Step.
Use cases
- Train custom LoRAs for specific characters, styles, or objects from a set of photos.
- Create edit LoRAs from before/after photo pairs to apply a consistent edit (e.g., relighting, color grade).
- Train music LoRAs on your own songs using ACE-Step, with previews playable in the gallery.
- Use Fizgig LoRA Isolation to train multiple people into one LoRA while keeping each person's face distinct.
- Repair broken or poorly trained LoRAs by adjusting individual blocks without retraining.
- Compare training epochs visually with LoRA Royale to select the best checkpoint.
Getting started
See the repository README for installation and quick-start instructions.
License and pricing
License: Apache-2.0. Pricing: open source and free.
Based on the project's public README and repository metadata.