Image-to-Image
Transforming an input image into a new image following a text prompt or style - starting diffusion from a noisy version of the input.
Think of it like
Photoshop with AI: you have a photo, you describe a change, the AI repaints parts of it.
Example
"A photo of a dog" -> "A painting of the same dog in Van Gogh style." Or: a room photo -> "Make it Victorian-era furniture."
How it actually works
Instead of starting diffusion from pure noise, encode the input image and add noise to it, then denoise. Strength parameter controls how much the model respects the input: 0 = no change, 1 = ignore input, generate fresh. Enables in-painting, outpainting, custom transformations. Requires the image and text encoder; reuses pretrained models. Allows fine-grained local editing if you mask regions.
For product teams
Powers generative editing tools (Photoshop AI, professional retouching).
For engineers
Requires VAE encoder to compress input, then add noise in latent space. Strength = diffusion timestep fraction.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome