These models allow editing and manipulating images in various ways. Key capabilities include:

• Inpainting & object removal - Remove objects or fill in missing image regions. Useful for removing unwanted elements from photos.
• Background removal - Delete and replace image backgrounds. Isolate subjects onto transparent or alternate backgrounds.
• Guided editing - Make targeted edits to images by adding guiding information like depth maps, sketches, edge detection, human poses, or text prompts. Allows fine-grained creative control.

Note: For face restoration and photo colorization, see our Upscale Images collection.

Our Picks

Best Inpainting Model: logerzhu/ad-inpaint

For removing objects and filling in missing regions of an image, we recommend starting with logerzhu/ad-inpaint. It delivers great results and runs faster than alternatives like instruct-pix2pix.

Best Background Removal: pollinations/modnet

To isolate a subject and replace the background of an image, pollinations/modnet is the clear choice. It’s optimized for this specific task and widely used.

Best Guided Editing: alaradirik/t2i-adapter-sdxl-depth-midas and other t2i-adapter models

For the most control over editing an image, try the alaradirik/t2i-adapter-sdxl-depth-midas model. It lets you modify an image using an automatically generated depth map.

The other models in the t2i-adapter-sdxl-* series are also great for guided editing using different types of inputs:

• alaradirik/t2i-adapter-sdxl-lineart - Edit using line art
• alaradirik/t2i-adapter-sdxl-canny - Edit using edge detection
• alaradirik/t2i-adapter-sdxl-sketch - Edit using sketches
• alaradirik/t2i-adapter-sdxl-openpose - Edit using human pose detection

Promising New Approaches

For even more advanced guided editing, two newer models are worth exploring:

adirik/masactrl-sdxl allows you to edit specific regions of an image using text prompts. It provides fine-grained control for local edits.

adirik/stylemc is a unique model that generates new images in the style of an input image based on a text prompt. While not a traditional editing model, it enables creative exploration.