Update README.md

README.md

@@ -7,7 +7,9 @@ Q4_0 quantized SDXL models in GGUF format, to be used with the custom ComfyUI no
 
 These are made with llama-quantize from [llama.cpp b3600](https://github.com/ggerganov/llama.cpp/releases/tag/b3600) patched with https://github.com/city96/ComfyUI-GGUF/blob/main/tools/lcpp.patch
 
-Why not Q4_K or higher quants? Having tested them all from Q3_K_S to Q8_0 for various finetunes, it became clear that only Q4_0 gives consistently good results for SDXL models, better than Q5, Q6 or even Q8. If the gens from Q4_0 aren't good enough for you then it's probably best to use the original safetensors unet and load the weights at FP8 instead of any GGUF quant.
+A couple of models are quantized with [llama.cpp b3962](https://github.com/ggerganov/llama.cpp/releases/tag/b3962) patched with lcpp_sd3.patch instead, which may give slightly different results than the default b3600-quantized ggufs but not necessarily better.
+
+Why not Q4_K or higher quants? Having tested a selection from Q3_K_S to Q8_0 for various finetunes, it seems that Q4_0 gives consistently good results for SDXL models, usually not worse than Q5, Q6 or even Q8. If the gens from Q4_0 aren't looking good enough, then it's usually best to use the original safetensors unet and load the weights at FP8 instead of any GGUF quant.
 
 https://huggingface.co/cyberdelia/CyberRealisticXL : https://civitai.com/models/312530?modelVersionId=1198781 <br>
 https://huggingface.co/RunDiffusion/Juggernaut-XI-v11 : https://civitai.com/models/133005?modelVersionId=782002 <br>