An image generated by ByteDance’s Seedream 5.5, released by the company to show its realism. Credit: ByteDance Seed (AI-generated with Seedream 5.5) ByteDance released Seedream 5.5, an AI image model it says keeps unchanged parts of an image intact over 100 rounds of edits and creates transparent-background assets. It is rolling out in ByteDance’s Chinese apps. The performance claims are ByteDance’s own.
ByteDance has released Seedream 5.5, a new AI image model, its Seed research team said on Monday. ByteDance says it can edit the same picture 100 times in a row without it falling apart. It is rolling out slowly in ByteDance’s Chinese apps, such as Doubao and Jimeng, and its Volcengine cloud. AI image editors often wear an image down when each output becomes the input for the next change.
ByteDance says Seedream 5.5 keeps the parts that should not change identical, down to the pixel, with no colour shifts. In one demo, it edited the same portrait 100 times, swapping clothes and objects while still adjusting small items like rings. The model can also create images with transparent backgrounds, so designers can drop elements such as a snowy owl straight into layouts without cutting them out by hand. ByteDance says it also renders skin, light and dense small text more realistically.
In ByteDance’s own tests, Seedream 5.5 beat OpenAI’s GPT Image 2.5 and Google’s Nano Banana 2.1 on text and on realism. Those results have not been checked independently, The Decoder noted. ByteDance says serious text errors appeared in 0.4% of its test images. Seedream sits alongside ByteDance’s Seedance video model.
Both are covered by a copyright pact ByteDance signed with Hollywood’s studios in August. A Pro version of Seedream 5.5 is coming, the company said. Get the TNW newsletter ✕ Close Get the TNW newsletter Get the most important tech news in your inbox each week.
