qwen-image-2.1-gguf

<Gallery />

able to use pig-clip and adapter as text encoder for fast speed or low vram alternative; optional: mmproj for image edit; tune the parameters for effect/details (see the first example below)

turbo (8-step)

ggk diffuser engine -- --diffusion-model qwen-image-2.1-turbo-nvfp4.gguf --vae pig_qwen_image_2.1_vae_fp32-f16.gguf --llm pig_clip-f16.gguf --llm-adapter pig_qwen3vl_8b_adapter-f16.gguf -p "cute anime girl with massive fennec ears and a big fluffy fox tail with long wavy blonde and pink hair between eyes and large blue eyes blonde colored eyelashes chubby wearing oversized clothes summer uniform long blue maxi skirt muddy clothes happy sitting on the side of the road in a run down dark gritty cyberpunk city with neon and a crumbling skyscraper in the rain at night while dipping her feet in a river of water she is holding a sign that says 'pig engine is the best! run it with ggk' written in cursive" --steps 8 --cfg-scale 1 --sampling-method euler -v --diffusion-fa --sigmas "1.0,0.978453,0.95418,0.926626,0.89508,0.845148,0.704534,0.414568,0.0" -o out.png

text to image

ggk diffuser engine -- --diffusion-model qwen-image-2.1-nvfp4.gguf --vae pig_qwen_image_2.1_vae_fp32-f16.gguf --llm pig_clip-f16.gguf --llm-adapter pig_qwen3vl_8b_adapter-f16.gguf -p "an anime pig holding a sign says GGUF" -W 512 -H 512 --sampling-method euler --diffusion-fa -v -o out.png

image edit

ggk diffuser engine -- --diffusion-model qwen-image-2.1-nvfp4.gguf --vae pig_qwen_image_2.1_vae_fp32-f16.gguf --llm pig_clip-f16.gguf --llm-adapter pig_qwen3vl_8b_adapter-f16.gguf --llm_vision mmproj-qwen3vl-8b-it-q8_0.gguf --ref-image sheep.png -p "a sheep in sunglasses" --cfg-scale 1.00 --steps 4 --sampling-method euler --diffusion-fa -v -o out.png

for running in comfyui, please see the demo above/below (model + encoder + vae; get workflow json or workflow-edit json)

supported clips (ggk and comfyui)

benchmark (clips)

replicate the results (benchmark table) above by:

ggk diffuser engine -- --diffusion-model qwen-image-2.1-nvfp4.gguf --vae pig_qwen_image_2.1_vae_bf16.gguf --llm pig_clip-f16.gguf --llm-adapter pig_qwen3vl_8b_adapter-f16.gguf --llm_vision mmproj-qwen3vl-8b-it-q8_0.gguf --ref-image sheep.png -p "a sheep in sunglasses" --cfg-scale 1.00 --steps 4 --sampling-method euler --diffusion-fa -v -o out.png

ggk diffuser engine -- --diffusion-model qwen-image-2.1-nvfp4.gguf --vae pig_qwen_image_2.1_vae_bf16.gguf --llm qwen3vl-8b-nvfp4.gguf --ref-image sheep.png -p "a sheep in sunglasses" --cfg-scale 1.00 --steps 4 --sampling-method euler --diffusion-fa -v -o out.png

ggk diffuser engine -- --diffusion-model qwen-image-2.1-nvfp4.gguf --vae pig_qwen_image_2.1_vae_bf16.gguf --llm qwen3vl-8b-it-q4_k_m.gguf --llm_vision mmproj-qwen3vl-8b-it-q8_0.gguf --ref-image sheep.png -p "a sheep in sunglasses" --cfg-scale 1.00 --steps 4 --sampling-method euler --diffusion-fa -v -o out.png

*tested with a basic 4050 laptop version gpu with 6GB vram, if you run it with a higher tier card, probably shorten the time to couple seconds but the gap still obvious