Volcengine released the Doubao image creation model Seedream5.5 on November 9, supporting "pixel-level controllable editing", multi-turn editing consistency, and native RGBA transparent asset generation. It launched on Volcengine Ark this week. Compared with the previous generation Seedream5.0Pro, the new model’s realism is closer to actual photography and can generate more natural, more cinematic images.
In terms of pricing, Seedream5.5 uses two-part billing based on "input image + output image". Text tokens are free. Editing a single 2K square image costs only 0.28 yuan, 60% cheaper than Seedream5.0Pro, and generation speed is nearly 40% faster. Developers and enterprise customers can integrate it with Harness and production pipelines, covering scenarios such as e-commerce marketing, advertising design, film and game assets, film storyboards, and knowledge visualization. Consumer users can experience it directly in Dreamina and Doubao Work.
The core of this upgrade is the pixel-level preservation capability of editing. After specialized training, the model is less prone to background drift, subject deformation, and image quality degradation during continuous modifications. In tests, even after hundreds of consecutive edits (with each round using the previous result as input), subject features, composition proportions, and lighting style outside the specified edit area remained completely consistent, achieving "change only where you edit".
