convertor/minimax-h3-image-gguf

Prompt

a lovely pig holding a sign says GGUF

Prompt

Picture 1

Prompt

Edit Picture 1. Put a pair of black sunglasses on the sheep, keep everything else unchanged.

Prompt

a lovely pig holding a sign says GGUF (with prompt upsampler)

Prompt

a lovely pig holding a sign says GGUF (with prompt upsampler)

text to image

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl_32b_minimax_h3-q2_k_m.gguf --vae minimax_h3_image_vae_fp16-f16.gguf  -p "a lovely pig holding a sign says GGUF" --steps 6 --cfg-scale 1 --offload-to-cpu -v --diffusion-fa --sigmas "1.0,0.9675236940383911,0.8875485062599182,0.75,0.5327739119529724,0.23200887441635132,0.0" -o out.png

image edit

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl_32b_minimax_h3-q2_k_m.gguf --vae minimax_h3_image_vae_fp16-f16.gguf -r sheep.png -p "Edit Picture 1. Put a pair of black sunglasses on the sheep, keep everything else unchanged." --steps 6 --cfg-scale 1 --offload-to-cpu --diffusion-fa -o out.png

text to image with prompt enhancer/upsampler (9b)

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl_32b_minimax_h3-q2_k_m.gguf --vae minimax_h3_image_vae_fp16-f16.gguf --prompt-upsampler qwen3.5-9b-q4_k_m.gguf -p "a lovely pig holding a sign says GGUF" --steps 6 --cfg-scale 1 --offload-to-cpu -v --diffusion-fa --sigmas "1.0,0.9675236940383911,0.8875485062599182,0.75,0.5327739119529724,0.23200887441635132,0.0" -o out.png

image edit with prompt enhancer/upsampler (9b + mmproj)

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl_32b_minimax_h3-q2_k_m.gguf --vae minimax_h3_image_vae_fp16-f16.gguf --prompt-upsampler qwen3.5-9b-q4_k_m.gguf --prompt-upsampler-vision mmproj-qwen3.5-9b-f16.gguf -r sheep.png -p "Edit Picture 1. Put a pair of black sunglasses on the sheep, keep everything else unchanged." --steps 6 --cfg-scale 1 --offload-to-cpu --diffusion-fa -o out.png

text to image with prompt enhancer/upsampler (4b)

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl_32b_minimax_h3-q2_k_m.gguf --vae minimax_h3_image_vae_fp16-f16.gguf --prompt-upsampler qwen3vl-4b-it-q4_k_m.gguf -p "a lovely pig holding a sign says GGUF" --steps 6 --cfg-scale 1 --offload-to-cpu -v --diffusion-fa --sigmas "1.0,0.9675236940383911,0.8875485062599182,0.75,0.5327739119529724,0.23200887441635132,0.0" -o out.png

image edit with prompt enhancer/upsampler (4b + mmproj)

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl_32b_minimax_h3-q2_k_m.gguf --vae minimax_h3_image_vae_fp16-f16.gguf --prompt-upsampler qwen3vl-4b-it-q4_k_m.gguf --prompt-upsampler-vision mmproj-qwen3vl-4b-it-f16.gguf -r sheep.png -p "Edit Picture 1. Put a pair of black sunglasses on the sheep, keep everything else unchanged." --steps 6 --cfg-scale 1 --offload-to-cpu --diffusion-fa -o out.png

*gk engine is now adopting an auto residency streaming (test) for machine doesn't has enough vram, it works for low tier gpu with even less than 6gb vram

opt to replace 32b with 4b + adapter

ggk --diffusion-model minimax_h3_image_pruned-nvfp4.gguf --llm qwen3vl-4b-it-q4_k_m.gguf --llm-adapter qwen3vl_4b_adapter-f16.gguf --vae minimax_h3_image_vae_fp16-f16.gguf -p "a lovely pig holding a sign says GGUF" --steps 6 --cfg-scale 1 --offload-to-cpu -v --diffusion-fa --sigmas "1.0,0.9675236940383911,0.8875485062599182,0.75,0.5327739119529724,0.23200887441635132,0.0" -o out.png

*p-h3-image works the same way (see below) in ggk as well

text to image

ggk --diffusion-model p-h3-image_w4a8.safetensors --llm qwen3vl-4b-it-q4_k_m.gguf --llm-adapter qwen3vl_4b_adapter-f16.gguf --vae minimax_h3_image_vae_fp16-f16.gguf -p "a lovely pig holding a sign says GGUF" --steps 6 --cfg-scale 1 --offload-to-cpu -v --diffusion-fa --sigmas "1.0,0.9675236940383911,0.8875485062599182,0.75,0.5327739119529724,0.23200887441635132,0.0" -o out.png

image edit

ggk --diffusion-model p-h3-image_w4a8.safetensors --llm qwen3vl-4b-it-q4_k_m.gguf --llm-adapter qwen3vl_4b_adapter-f16.gguf --llm_vision mmproj-qwen3vl-4b-it-f16.gguf --vae minimax_h3_image_vae_fp16-f16.gguf -r sheep.png -p "Edit Picture 1. Put a pair of black sunglasses on the sheep, keep everything else unchanged." --steps 6 --cfg-scale 1 --offload-to-cpu --diffusion-fa -o out.png

*ggk supports both int8_convrot and w4a8 file formats right away