gguf-org/mageflow-gguf overview

---

license: mit

base_model:

• microsoft/Mage-Flow

widget:

• text: >-

change the cat to a dog

output:

url: dog.png

• text: >-

change the dog to a cat

output:

url: cat.png

• text: >-

add a headset to the sheep

output:

url: headset.png

---

mageflow-gguf

• edit-turbo

<Gallery />

run it with gguf-diffusion

gguf-diffusion

!screenshot

or run it with gguf-connector

ggc fu

benchmark (initial comparisons)

• inference (engine)

| |ggk (gk) |gguf-cpp/gguf-diffusion (ggml kernel) |stable-diffusion.cpp (full ggml)

|------------|-------------------|----------------|--------------------------|

|completed in |7-8s |24-25s |25-26s

|speed (base/complete) |3x |base (1x) |.95-1x

• replicate the results (benchmark table) above by:

ggk diffuser engine -- --diffusion-model mageflow-edit-turbo-nvfp4.gguf --vae pig_mageflow_vae_fp32-f16.gguf --llm qwen3vl-4b-it-q4_k_m.gguf --llm_vision mmproj-qwen3vl-4b-it-f16.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

gguf-cpp diffuser engine -- --diffusion-model mageflow-edit-turbo-nvfp4.gguf --vae pig_mageflow_vae_fp32-f16.gguf --llm qwen3vl-4b-it-q4_k_m.gguf --llm_vision mmproj-qwen3vl-4b-it-f16.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

./sd-cli --diffusion-model mageflow-edit-turbo-nvfp4.gguf --vae pig_mageflow_vae_fp32-f16.gguf --llm qwen3vl-4b-it-q4_k_m.gguf --llm_vision mmproj-qwen3vl-4b-it-f16.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