Seedream 5.0 Pro

Seedream 5.0 Pro is the flagship image model from ByteDance Seed, the AI research group at the company behind TikTok. It was first shown at ByteDance's Volcano Engine FORCE conference in Beijing on 23 June 2026 and launched on 8 July. ByteDance pitches it at professional production work, where an image has to match what a designer had in mind rather than just look attractive.

What it's for

The model is built for pictures packed with information, such as infographics, charts and pages that mix text and images. It can write text directly into images in more than ten languages. ByteDance also says it has improved realistic photos and portraits, aiming to combine the polish of computer graphics with the look of a real photograph.

Editing by pointing

Its biggest change is in editing. Instead of describing every change in words, you can click on a spot or draw a loop around an area and change just that part, swap a colour or a material, or turn a rough sketch into a finished image. It can also split a picture into separate layers, with the background and each element apart, so they can be moved or reused. It can combine several images into one and takes up to fourteen reference pictures.

How it differs from Lite

Seedream 5.0 Lite, released in February 2026, is the cheaper, faster tier, and the one that reasons about a request and searches the web. Pro is the production-design tier that adds the point-and-click editing, layer splitting and infographic features. The Seedream line itself goes back to late 2024 and has been known from the start for handling Chinese and English prompts and lettering well.

Where you'll see it

Seedream 5.0 Pro is built into the AI image tools in CapCut, ByteDance's popular video editor, and into its Dreamina creative platform. Businesses can use it through BytePlus, ByteDance's international cloud service, and through resellers such as OpenRouter. ByteDance has not published benchmark scores for it, and admits there is still room to improve in very fine text and in keeping pixel-level edits consistent.