Image Editing
Modifying an image by describing changes in natural language - remove an object, change a color, add elements.
Think of it like
Photoshop but with text instructions: "remove the sign" instead of using the eraser tool.
Example
"A cat wearing sunglasses" -> modify the original cat photo to add sunglasses. "Change the sky to purple" -> repaints the sky.
How it actually works
Inpainting: mask the region to edit, then diffusion fills it. Guided editing: text-guided diffusion + masking. Challenges: semantic consistency (edited region matches surroundings), identity preservation (original object recognizable). Methods: diffusion + mask, optimization-based (LPIPS loss), or learned editing networks. Requires careful balance of adherence to prompt vs. realism.
For product teams
Consumer photo editing, professional retouching, content creation.
For engineers
Text encoder + image encoder + diffusion UNet + masking. Loss: reconstruction (original) + guidance (text).
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