One Obsession v24 Quant
Diffusion-model-only quantization of One Obsession v24 for ComfyUI. It uses a mixed plain INT8 and ConvRot INT8 format with ConvRot group size 256.
The CLIP-L, CLIP-G, and VAE remain in their original FP16 form. This is an experimental storage-compression quant.
Link to original model by maxfeifei8
Test Settings
- 25 Steps
- CFG 4.5
- Euler Ancestral / Normall
- 696 × 1048 resolution
Observations
- Noted about 45% speed
- Quality appears good enough for my untrained eye
Hardware Testing
| Component | Specification |
|---|---|
| GPU | NVIDIA GTX 1660 Super 6GB |
| RAM | 32GB |
Sample Images
