I've done some work to convert Krea2 to GGUFs. It should come in handy for GPUs with <8GB VRAM.
Get the GGUF loader nodes from here: https://github.com/molbal/ComfyUI-GGUF
Get the converted models from here: https://huggingface.co/molbal/krea2-gguf
Q4_0 uses 7573MB on my GPU, but GGUFs are generally easier to offload if you have less VRAM.
Description
FAQ
Comments (16)
can you please convert this RedCraft Krea2 to Q4 gguf please ; RedCraft | 红潮 | KREA 2 赤佬 Red Mix Edition NSFW - Krea2 赤佬 Edition | Krea 2 Checkpoint | Civitai
can you please convert this Krea2 to Q8_0 gguf please.https://civitai.red/models/2731187/moody-krea-2-mix-uncensored
@molbal Thank you very much!
Comfy don't seem to recognize Krea2 architecture on GGUF yet
Yeah I forked the nodes and patched them to recognize Ideogram 4 and Krea 2: https://github.com/molbal/ComfyUI-GGUF
bless you
Thanks for the nodes! Is it possible to upload the raw models to Civitai too? Hugging Face is working very badly for me at the moment. Thank you!
Ah yes I forgot. Quite the opposite for me, I can use Hugging Face 1Gbps for both upload and download, but Civitai is much slower. I'll upload them and ping you here
can you please convert this RedCraft Krea2 to Q4 gguf pleasehttps://civitai.red/models/2735032/pornmaster-krea2?modelVersionId=3119653
It's still in early access, but ping me again when it is possible to download.
If you already have it downloaded and the custom nodes installed, then you can run this script to convert it yourself: https://github.com/molbal/ComfyUI-GGUF/blob/main/tools/convert_krea2_gguf.py
1) You have miss typing in your note at Github repo: "To use it, clone https://github.com/city96/ComfyUI-GGUFand not the original repo." <--- original repo adress, except your repo adress
2) I can't load any CLIP GGUF. All attempts cause WARINING errors. Enable Issue option on your Github repo. Even your workflow from first image, with exactly the same CLIP model - Qwen3VL-4B-Instruct-Q4_K_M.gguf - cause 'WARNING: clip missing'... Wtf...
I spent a day specifically testing GGUF, and this checkpoint is the best. The body size is well controlled, and the character expressions are rich and closer to natural faces. Finally, I recommend that friends with 8G choose the Q5_0 version without hesitation.
Thank you very much.
I do not want to gatekeep, this is my script for conversion: https://github.com/molbal/ComfyUI-GGUF/blob/main/tools/convert_krea2_gguf.py