JoyAI-Image¶
| Model | Source | Revision | Weight license | Parity (CPU, fp32) |
|---|---|---|---|---|
joyai-image-edit (+ image input) |
jdopensource/JoyAI-Image-Edit-Diffusers | 4b41fb2 |
Apache-2.0 | Verified |
joyai-image-edit-plus (+ image input) |
jdopensource/JoyAI-Image-Edit-Plus-Diffusers | c268646 |
Apache-2.0 | Verified |
Catalog: joyai-image-edit ·
Models: joyai-image-edit, joyai-image-edit-plus · Task: t2i
Architecture. JoyAI-Image combines Qwen3-VL conditioning, a WAN causal VAE, a dual-stream 3D transformer, and flow-match Euler sampling. Published weights are bfloat16 safetensors.
Settings. steps (40 for Edit, 30 for Edit Plus), guidance (4.0), width/height, seed,
and image. The base checkpoint also permits text-only generation; Plus requires an input image.
Validation. Published-header and exact random-weight tests cover the native Qwen3-VL vision conditioning and both transformer paths. Real-checkpoint parity is covered: all 23 real tests passed with exactly 0 difference in every recorded comparison, and the run took 123 minutes on an M2 Max (CPU) (validation). The catalog records the exact repositories, commits, component files, and bfloat16 precision.