gguf quantized version of mochi (test pack for gguf-node)

setup (once)

• drag mochi-q3_k_m.gguf [4.31GB] to > ./ComfyUI/models/diffusion_models
• drag t5xxl_fp16-q4_0.gguf [2.9GB] to > ./ComfyUI/models/text_encoders
• drag mochi_vae_fp8_e4m3fn.safetensors [460MB] to > ./ComfyUI/models/vae

run it straight (no installation needed way)

• run the .bat file in the main directory (assuming you are using the gguf-node pack below)
• drag the workflow json file (below) to > your browser

workflow

• example workflow (with gguf encoder)
• example workflow (safetensors)

review

• revised workflow to bypass oom issue and around 50% faster with the new fp8_e4m3fn file
• t5xxl works fine as text encoder; more quantized versions of t5xxl can be found here
• gguf with pig architecture is working right away; welcome to test

reference

• base model from genmo
• pig architecture from connector
• comfyui from comfyanonymous
• gguf-node (pypi|repo|pack)

prompt test#

prompt: "a fox moving quickly in a beautiful winter scenery nature trees sunset tracking camera"