Note: For more advanced configurations, see the tutorial, dataloader configuration guide, and the options breakdown pages.
For the complete and most accurate feature matrix, please see the main README.md.
| Model | Params | PEFT LoRA | Lycoris LoKr | Full-Rank | Quantization | Mixed Precision | Grad Checkpoint | Flow Shift | ControlNet LoRA |
|---|---|---|---|---|---|---|---|---|---|
| PixArt Sigma | 0.6B-0.9B | ✓ | ✓ | optional (int8) | bf16 | ✓ | ✓ | ||
| NVLabs Sana | 1.6B-4.8B | ✓ | ✓ | optional (int8) | bf16 | ✓+ | ✓ | ||
| Kwai Kolors | 2.7 B | ✓ | ✓ | ✓ | not recommended | bf16 | ✓ | ||
| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | ✓ | optional (int8, fp8, nf4) | bf16 | ✓+ | ✓ (SLG) | ✓ |
| Flux.1 | ~8B-12B | ✓ | ✓ | ✓* | optional (int8, fp8, nf4) | bf16 | ✓+ | ✓ | ✓ |
| Chroma 1 | 8.9B | ✓ | ✓ | ✓* | optional (int8, fp8, nf4) | bf16 | ✓+ | ✓ | |
| Auraflow | 6 B | ✓ | ✓ | ✓* | optional (int8, fp8, nf4) | bf16 | ✓+ | ✓ (SLG) | ✓ |
| HiDream I1 | 17 B (8.5B MoE) | ✓ | ✓ | ✓* | optional (int8, fp8, nf4) | bf16 | ✓ | ✓ | ✓ |
| OmniGen | 3.8 B | ✓ | ✓ | ✓ | optional (int8, fp8) | bf16 | ✓ | ✓ | |
| Stable Diffusion XL | 2.6 B | ✓ | ✓ | ✓ | not recommended | bf16 | ✓ | ✓ | |
| Lumina2 | 2B | ✓ | ✓ | ✓ | optional (int8) | bf16 | ✓ | ✓ | |
| Cosmos2 | 2B | ✓ | ✓ | ✓ | not recommended | bf16 | ✓ | ✓ | |
| LTX Video | ~2.5 B | ✓ | ✓ | ✓ | optional (int8, fp8) | bf16 | ✓ | ✓ | |
| Wan 2.x | 1.3B-14B | ✓ | ✓ | ✓* | optional (int8) | bf16 | ✓ | ✓ | |
| Qwen Image | 20B | ✓ | ✓ | ✓* | required (int8, nf4) | bf16 | ✓ (required) | ✓ | |
| Qwen Image Edit | 20B | ✓ | ✓ | ✓* | required (int8, nf4) | bf16 | ✓ (required) | ✓ |
Note: The above table provides a simplified overview. For the complete and most accurate feature matrix with detailed specifications, please see the main README.md.
ℹ️ The Wan quickstart covers 2.1 training plus the 2.2 high/low stage presets and the new time-embedding compatibility toggle.
⚠️ These tutorials are a work-in-progress. They contain full end-to-end instructions for a basic training session.