Seedream 5 Layered Image Editing
Seedream 5 decomposes a scene into transparent PNG layers and edits specific regions, with native text generation across fourteen languages.
- ByteDance
- 5 variants
- From $0.035 per image
- Price
- See price
- Speed
- ~1 min
- Best for
- Image edit · Image to Image
- Price
- See price
- Speed
- ~3 min
- Best for
- Image edit · Image to Image
- Price
- See price
- Speed
- ~3 min
- Best for
- Text to Image
- Price
- See price
- Speed
- ~30s
- Best for
- Text to Image
- Price
- See price
- Speed
- ~45s
- Best for
- Image edit · Image to Image
More from ByteDance.
A family groups the versions of one model line. Open one to compare its variants side by side.
Working with Seedream 5.
Seedream 5 is the generation that made ByteDance's image line layer-aware. Layer Decomposition splits a single scene into separate transparent PNGs a designer can actually edit; the Pro edit variant modifies a chosen region while leaving the rest untouched and supports sketch completion; and Pro text-to-image renders native text in fourteen languages.
When to choose Seedream 5
Choose Seedream 5 when the output has to be editable afterwards — ad layouts, product visuals, anything a designer will open in a layer editor. Seedream 4.5 is the pick when a single flat image with strong typography is enough.
Choosing between the variants
Layer Decomposition is the one that returns editable assets rather than a picture. Pro Edit is for precise regional changes, Pro Text to Image for complex layouts and multilingual type, and the Lite variants for 4K generation and plain-language edits at lower cost.
Questions about Seedream 5
The short answers. For anything specific to your setup, talk to our team.
Talk to an engineerPlan for your expected volume.A set of transparent PNG layers taken from one scene, so elements can be moved, replaced or restyled without regenerating the whole image.
Fourteen, natively, on the Pro text-to-image variant — which is what makes it usable for non-English campaign work.
Lite reasons about the prompt and can use real-time web search before generating, which is unusual for an image model and helps with factual subjects.