ByteDance released Seedream 5.5, its latest image-generation model, on November 9. The update adds support for up to 100 editing rounds, native transparent image output in RGBA format, and improved text rendering.

The model is available in Doubao and Doubao Workspace, while Jimeng AI and Volcano Ark are gradually opening trial access. ByteDance also plans to launch a Pro version later.

Republished tests found the model handled repeated edits and indoor scenes fairly stably. But it still made errors in human anatomy and small objects such as glasses.

The report said API editing for a 2K square image cost about 0.28 yuan (US$0.042) per image, less than half the price of Seedream 5.0 Pro.

Seedream 3.0 was benchmarked on fidelity, prompt alignment, and editing, while Seedream 4.0 improved controllability and tooling features.

China’s AI-content labeling rules, which cover generated and edited images, took effect in 2025.

🔗 Source: 36Kr