ideogram / remove-background

Overview

Remove Background by Ideogram extracts the main subject from an image and returns it with a real alpha channel. It uses a generative approach to rebuild edge pixels, which helps with hair, glass, reflections, and thin typography strokes. You get a transparent PNG cutout that’s ready to place on any background. This saves manual masking work and reduces halo and fringing artifacts.

What you can build

• E-commerce product cutouts for marketplaces, catalogs, and ads
• Profile photos and portraits with clean transparency for layouts
• Logo and typography cutouts for brand systems and templates
• Sticker packs and die-cut assets with smooth edges
• Batch prep pipelines for design teams and creative operations

Inputs

• One source image you want to cut out, uploaded as a single file.
• JPEG, PNG, and WebP work best. Other formats may be converted before processing.
• Keep the file under 10 MB. Larger files may be compressed to fit.
• If you upload more than one image, the run uses the first one.

Outputs

• One PNG image with transparency (alpha), containing only the foreground subject.
• A structured response that includes a hosted file reference for the PNG.
• A safety flag that indicates whether the returned image passed Ideogram’s safety checks.
• The hosted link is time-limited. Download the PNG if you need to keep it.

Limitations

• Input size is capped at 10 MB, and only JPEG, PNG, and WebP are supported reliably.
• The model returns a single cutout per run.
• Very low-resolution images can produce soft edges and missing fine detail.
• Motion blur, heavy noise, and compression artifacts can reduce edge quality.
• Subjects that match the background color closely can cause small holes or edge loss.
• Scanned images, screenshots with heavy artifacts, and poorly lit photos can confuse foreground detection.

Safety & compliance

• Check the returned safety indicator before you store or display results.
• Only process images you have rights to use and edit.
• Don’t use cutouts to mislead viewers (for example, deceptive composites of real people).
• Follow Ideogram’s content rules for any image you upload and redistribute.