ByteDance released the Seedream 5.5 image model on November 9, supporting 100 rounds of continuous editing and native transparent backgrounds
The model keeps non edited areas pixel stable and outputs images with an alpha channel directly, cutting severe text corruption to 0.4 percent, with a 2K square image edit costing about 0.28 yuan, less than half the previous version
On November 9, the ByteDance Seed team released the next generation image creation model Seedream 5.5 with three main capability upgrades. More stable multi-round editing keeps non edited regions pixel stable, removes color shift and reduces detail loss and quality decay during consecutive edits, supporting up to 100 rounds of continued creation around the same piece of work.
Second is native transparent asset generation, where the model directly outputs images with an alpha channel, handling contour edges and transparent regions finely and removing the need for background removal and retouching, so generated elements can enter later design work as standalone assets. Third is more natural texture and clearer text, improving lighting, materials and skin rendering, reducing excessive gloss and overexposure, and materially improving text rendering.
On evaluation, the team ran targeted tests across six scenarios covering e-commerce, marketing, film and television, gaming, office work and apparel design, where usability in the film and office scenarios more than doubled against the previous version, severe text corruption fell to 0.4 percent and light text corruption to 1.7 percent. Pricing keeps the two part structure of charging for the input image plus the output image, with text tokens costing nothing, so a 2K square image edit runs about 0.28 yuan per image, less than half the previous version. The model is rolling out across Jimeng, Doubao, Doubao Work and the Volcano Ark platforms.