Fizgig 7 - Anima and SDXL arrive, fine-tuning for every model & the community can now add support for new/old models
retained text
**Anima and SDXL join** using the new model driver system, with LoRA and LoKR training, sliders, fine-tuning and the full workbench (Repair Studio, LoRA the Explorer, Profiler, Extract, LoRA Royale). SDXL takes any single-file checkpoint; Juggernaut XL v9 is the default. **Full fine-tuning on every model:** Klein 9B, Krea 2, Qwen Image 2.1, MiniMax H3, SDXL and Anima. Every model fine-tunes on a 16 GB card, Anima and SDXL from 8 GB. On a bigger card it trains the whole model at once instead of a part at a time, which got through the model about 2.6x faster in my tests. | Model | Fine-tunes from | Whole model at once from | |---|---|---| | Anima | 8 GB | 10 GB | | SDXL | 8 GB | 10 GB | | Qwen Image 2.1 | 12 GB | 24 GB | | Klein 9B | 16 GB | 32 GB | | Krea 2 | 16 GB | 48 GB | | MiniMax H3 | 16 GB | 48 GB | Turn any fine-tune into a LoRA with Checkpoint to LoRA. (This gives you a Lora more capable than trained directly to Lora) **Add your own model.** The driver system is complete: every model now runs on it, and the guide I promised last time is in the repo (docs/drivers). SDXL and Anima were built by following it. PRs are welcome for new models and older favourites alike, and fine-tuning is optional: a model can ship without it and gain it later. [Fizgig on GitHub](https://github.com/shootthesound/Fizgig) · [7.0 release notes](https://github.com/shootthesound/Fizgig/releases/tag/v7.0.0) · [Add your own model](https://github.com/shootthesound/Fizgig/tree/master/docs/drivers)
document fetched 2026-10-06 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1wyfwdc/fizgig_7_anima_and_sdxl_arrive_finetuning_for