Kroma INT8 Quants
Experimental INT8 quantizations of Kroma (by lodestones), primarily targeting lower-VRAM and older consumer GPUs.
Available Versions
| Version | File Name | File Size |
|---|---|---|
| Mixed v3 | kroma_v0.2_turbo_mixed_v3.safetensors |
8.67 GB |
| Mixed v4 | kroma_v0.2_turbo_mixed_v4.safetensors |
8.01 GB |
| Mixed v5 | kroma_v0.2_turbo_mixed_v5.safetensors |
7.64 GB |
Version Differences
| Version | Description |
|---|---|
| v3 | More conservative profile prioritizing quality retention. |
| v4 | Reduced model size further while maintaining similar quality. |
| v5 | Most aggressive compression profile with the smallest file size. |
Hardware Testing
Tested on:
| Component | Specification |
|---|---|
| GPU | NVIDIA GTX 1660 Super 6GB |
| RAM | 32GB |
| Runtime | ComfyUI + Comfy Kitchen |
Recommendation
- Use v4 for the best balance between size and quality.
- Use v5 if minimizing storage usage is the priority.
- Use v3 if you prefer the most conservative profile.
Model tree for PotatoForge/Kroma-INT8-Quants
Base model
lodestones/Kroma