The three trainers most used for current image models, compared by model support (from each project's README) and activity (GitHub API).
| Model | ai-toolkit | OneTrainer | musubi-tuner |
|---|---|---|---|
| Fibo 2.0 / Edit Turbo (Bria, 8B, CC BY-NC 4.0) | unknown (no Fibo 2.0 support or PR yet, Oct 21) | unknown (no Fibo 2.0 support or PR yet, Oct 21) | unknown (no Fibo 2.0 support or PR yet, Oct 21) |
| FLUX.3 Image Klein (BFL) | supported (day-0, PR #1078 merged Oct 15; fixes Oct 15) | unknown | unknown |
| MOSS Image (OpenMOSS, ~5B, Apache 2.0) | supported (SFT + Base, Oct 16; edit training Oct 17) | unknown | unknown |
| Iris 3B (pixel-space, Apache 2.0) | supported (Oct 9) | unknown | unknown |
| Qwen-Image-2.1 / 2.1 Turbo | supported (Sep 20; Turbo via training adapter Oct 10) | unknown | unknown |
| Ming-Image-0.1-Design (inclusionAI, MIT) | supported (fine-tune, Sep 23; training adapter by default) | unknown | unknown |
| Krea 2 / Krea 2 Turbo | supported (Raw + Turbo, incl. edit training) | supported (Krea 2 16 GB preset benchmarked in OneTrainer PR #1680; not in README list) | supported (fp8 / ConvRot int8 base) |
| FLUX.2 dev (32B) | supported (reference) | supported | supported |
| FLUX.2 klein (4B) | supported (klein base 4B and 9B) | supported | supported |
| Z-Image | supported (Z-Image, Turbo, De-Turbo) | supported | supported |
| SDXL | supported | supported | not listed |
| FLUX.1 | supported | supported | FLUX.1 Kontext only |
| Trainer | Stars | Last commit | Latest release |
|---|---|---|---|
| ai-toolkit | 12373 | 2026-10-19 | none published |
| OneTrainer | 3247 | 2026-10-19 | none published |
| musubi-tuner | 2107 | 2026-10-12 | v0.3.6 |
Sources: ai-toolkit, OneTrainer, musubi-tuner, sanj.dev comparison.
Data updated 2026-10-20. RunPod and Vast.ai links are referral links. unknown = no data yet. · Raw data (JSON) · llms.txt