Image models
Seedream 5.5: native transparent images and 100-round editing
ByteDance Seed's Seedream 5.5 generates images with a real alpha channel and keeps unedited areas intact across up to 100 successive edits. For game artists, that means cut-out assets without a background-removal step.
ByteDance’s Seed team introduced Seedream 5.5 on 9 November 2026, a new image model aimed at professional production rather than one-off pictures. It brings three upgrades: steadier multi-turn editing, native transparent output, and more realistic images with clearer text.
Editing that does not drift
Image editors tend to degrade when one output becomes the next input: small errors pile up into blur and colour shifts. Seed says Seedream 5.5 was trained to preserve unedited regions at the pixel level and to avoid colour shifts in them, so work on one image can continue across 100 rounds of editing.
Its demo applies 100 successive edits to a single portrait, changing clothing, accessories and tabletop objects, while still making precise changes to small items such as rings and glasses frames. Another sequence turns an old study into a colourful room, moving a box around and adjusting perspective, occlusion and cast shadows at each step.
Native RGBA output
The model can generate images with an alpha channel, including full and partial transparency, directly from a text description or reference images. Seed says edges and transparent areas are rendered carefully, so assets can go into layouts and compositing “without separate background removal or edge cleanup”. The demos include a snowy owl with separated feathers, lettering styled from a reference texture, and character sets reused as references across scenes.
Realism and text
Seed says lighting, materials and skin textures look more natural, with less gloss and overexposure. Small text is improved too: in its Chinese and English tests, the severe text corruption rate fell to 0.4% and the minor rate to 1.7%.
The numbers, and their limits
All figures are Seed’s own. On the MagicArena platform, Seed tested six use cases: e-commerce, marketing, film and television, gaming, office productivity and fashion design. Seedream 5.5 improved both quality scores and usability rates over Seedream 5.0 Pro in all six, with usability more than doubling in film and television and in office productivity. In pairwise preference tests it scored +55.62% for image-to-image realism against the previous version, and Seed says it leads GPT Image 2.5 Sunburst and Nano Banana 2.1 on text structure.
The announcement does not give pricing, resolution limits or API terms. The project page carries “Try now” and “Get API” links, but we have not confirmed what is open to which regions. A separate BytePlus Lumina page still describes 5.5 as a preview, so check availability before planning around it. Seed also says a Pro version of the 5.5 series will follow.
Our take
For anyone building interactive worlds, native transparency is the practical part. Sprites, props, UI elements and character cut-outs usually need a background-removal pass that leaves fringes; generating them with an alpha channel removes that step. Stable long edit chains matter just as much, because a character or location is refined many times before it is right. The caveat is consistency across a whole cast and world, which a single model’s editing demo does not prove. How tools like this fit a creative workflow is covered in authorship in AI interactive storytelling.