Idle

Model

required
Select a .gguf / .safetensors model…

--diffusion-model for a standalone diffusion model file (Flux, SD3.5, Wan…); --model for an all-in-one checkpoint (SD1.x/SDXL).

--vae
Optional VAE…
--tokenizer-pack
Optional tokenizer pack folder…

Text Encoders

optional

No external text encoders. Needed for Flux / SD3.5 / Wan-style models (--clip_l, --t5xxl, --llm, …).

Additional Models

optional

No additional models (ControlNet, TAESD, upscaler, PhotoMaker, …).

Image Inputs

optional▾

Feed an initial, mask, end, control, or reference image — for img2img, inpainting, first/last-frame video, ControlNet guidance, or Kontext-style editing.

--init-img
Initial image…
--mask
Mask image…
--end-img
End-frame image (flf2v)…
--control-image
Control image…

Reference images --ref-image

Prompt

Generation

1.0 for turbo/flow models, 7.0 for standard SD.

4 for turbo/flow models, 20–30 for standard SD.

Flags

Low VRAM / memory (off by default)

Tensor Split (multi-GPU)

optional▾

Assign more than one device to a module to split it across GPUs — --backend --split-mode. Needs a build with multiple GPU backends (CUDA/Vulkan/ROCm); layer split spreads whole transformer blocks per device, row split shards individual weights (CUDA only).

Output

directory
Default: Downloads folder

Images are written as gguf-<timestamp>.png (batches: …_0000.png, …_0001.png, …).

Command


        
        
      


      
      

        
        
      

    


    
    

    
    

    
    

    
    

  


  
  

    (no model)
    |4 steps · cfg 1
    |512×512
    |euler
    
    Idle