ByteDance releases Seedream 5.5: continuous edits across 100 rounds on a single image, and can directly generate transparent assets

Beating AI News Flash: ByteDance releases its next-generation image model Seedream 5.5, supporting up to 100 consecutive rounds of edits to the same image, and it can also directly generate transparent-background assets.

ByteDance demonstrated a 100-round continuous editing session. Centered on the same character, each round was based on the image from the previous round, continuously changing clothing, glasses, rings, and desktop items. After repeated edits, the character still retained its original features, and the model could also accurately adjust small objects such as rings and frames. The company says the new version can keep unedited areas pixel-identical, avoiding the problems of past iterations where repeated edits caused increasing distortion and color drift.

Another new feature is native RGBA transparent asset generation. For example, asking the AI to draw a snowy owl with its wings spread produces a result with a transparent background built in, and it can even handle the tiny gaps between feathers. Stickers, artistic text, and product assets can all be placed directly into posters or design drafts, eliminating the manual cutout step.

Generation quality has also improved. The new version reduces issues such as overly smooth portrait skin, glossy-looking images, and localized overexposure, and it can generate more complex Chinese text and dense small text. According to ByteDance's official tests, the severe text corruption rate dropped to 0.4%, and the minor text corruption rate is 1.7%. The usability rate of generated results for film and television and office scenarios has reached more than double that of the previous generation, Seedream 5.0 Pro.

Seedream 5.5 is gradually rolling out on Jimeng, Doubao, Doubao Work, and Volcano Ark. A Pro version will also be launched later.