An anime/comics style Flux checkpoint.
Note: The "ONNX" fp8 model isn't ONNX at all, but a Q5_K_M quantized GGUF model. If your computer is a potato, yam, or some other type of tuber, you may have a better chance to get that one to run, and the quality hit isn't too bad.
Description
FAQ
Comments (2)
Looks quite good also want to especially thank you for making a Q5_K_M. Everyone's making FP8's and I feel like it's just worse + requiring more VRAM compared to Q5_K_M. If I really loved a model I could in theory quantize it myself but because they only release the FP8's, it would be double-quantized (and therefore poorer quality) than going straight from FP16 to Q5_K_M. IIRC city96 is the one that has quantization code for this right? I know there's also leejet's stable-diffusion.cpp, but I don't see K-quant options there.
The tool I used to go from FP16-safetensors to FP16-gguf to Q5_K_M-gguf is here:
https://github.com/city96/ComfyUI-GGUF/tree/main/tools
It actually comes with Comfy's GGUF module, although there are some additional installation steps (which are detailed on that page).
I'm a big fan of Q5_K_M... In my experiments with Llama finetunes, it seems like a great balance between quality and size. Quality starts to drop off significantly after that.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.