Model Support Overview - Nerogar/OneTrainer GitHub Wiki
This overview aims to list the supported models in OT, available training types and token limit. Check their pages for more detailed information, training resolution and architecture. Almost all combinations has a default preset.
Inference support means the support for models trained with OT, considered inferences are Comfy/Swarm, Forge (see notes) and OT sampling tool.
| Model | Fine Tune | Lora | Embedding | Max Token | Inference Support |
|---|---|---|---|---|---|
| SD (1.5 to 3.5, SDXL) | X | X | X | 75 | All |
| Flux.1-Dev | X | X | X | 75 | Lora and FT: All, Embedding: OT only |
| Chroma | X | X | X | 512 | Lora and FT: All, Embedding: OT only |
| Qwen | X | X | - | 512 | All |
| Hidream | X | X | X | 120 | OT only |
| Hunyuan video | X | X | X | 75 | OT only |
| Pixart alpha | X | X | X | 120 | All |
| Pixart sigma | X | X | X | 120 | All |
| Sana 1.6b | X | X | X | 300 | All |
| Stable Cascade | X | X | X | 75 | All |
| wuerstchen 2.0 | X | X | X | 75 | All |
Notes:
- HiDream Token limit are according to the code 120 for all four text encoders. As CLIP doesn't support 120, it might be a bug.
- There might be some exceptions on the old models like Cascade
- Forge is a surviving project, abandonned by his author but maintained by supporters and has several forks. Its support is just indicative and depends on the fork.