Use Sketch to Guide Image Generation with Structure and Shape

Sketch lets you guide image generation using the structure of a reference image.

Instead of describing layout and pose in words, you can upload an image, extract a sketch from it, and use it as the base for generation. The result will follow that sketch — more or less, depending on the strength setting you choose.

How to use it

1. Click Reference image in the prompt panel  

2. Set Mode = Reference  

3. Choose Method = Sketch  

4. Set the strength (default is 3)  

5. Click Create Image

How it works

The uploaded image is converted into a sketch, which is then used to guide the shape and structure of the generated image  

Sketch strength ranges from 1 (very loose interpretation) to 5 (almost exact copy)  

Higher strength values will keep the layout of the sketch more strictly, which is helpful if your sketch includes important details  

If the image used for the sketch has incorrect proportions, strange lines, or anatomy that doesn’t match the model’s training, it may override the model’s intended style and lead to distorted results — especially at high strength  

Your prompt should match the sketch. For example, if the sketch shows a tree but the prompt says “a cat,” the output might look confusing  

We recommend using a reference image with the same aspect ratio as the final output.  

If the proportions don’t match, the sketch will be scaled and fitted into the frame while keeping its original aspect ratio — which might change the composition.

Model compatibility

✅ Sketch works only with style models.

⚠️ It is not supported by:

• Flux.1 Kontext  

• Flux Dev  

• GPT Image  

• Imagen 4 Ultra  

• Juggernaut Flux Pro