sd3.5-lite-gguf

• use 8-step which is faster around 70% than the original 24-step; quality remains (it depends)
• run it with gguf-connector; simply execute the command below in console/terminal

ggc s3

GGUF file(s) available. Select which one to use:

1. sd3.5-2b-lite-iq4_nl.gguf
2. sd3.5-2b-lite-iq4_xs.gguf
3. sd3.5-2b-lite-mxfp4_moe.gguf

Enter your choice (1 to 3): _

• select gguf in your currently working directory to interact with; then start your first prompt
• you will get the image output in few seconds (2b) even with the beginner level gpu or cpu

run it with gguf-node via comfyui

• drag sd3.5-lite to > ./ComfyUI/models/diffusion_models
• t5xxl [2.75GB], clip-g [1.39GB], clip-l-v2 [248MB] to > ./ComfyUI/models/text_encoders
• drag pig [168MB] to > ./ComfyUI/models/vae

reference

• base models from stabilityai
• for model merge details, see ariG23498 and sayakpaul
• comfyui from comfyanonymous
• gguf-node (pypi|repo|pack)
• gguf-connector (pypi)