ByteDance Ships Seedream 5.0 Pro: Circle a Region, It Edits There
TL;DR
ByteDance Seed launches Seedream 5.0 Pro with interactive point-and-edit, native rendering in 14 languages, and output pricing from 0.3 CNY per image.
ByteDance's Seed team released Seedream 5.0 Pro on the evening of July 8, 2026, a multimodal image model that pushes four upgrades: interactive precision editing, native multilingual generation, realistic imagery, and dense-information visualization. The model is live on the Volcano Ark playground and rolls out to Doubao and Jimeng next.
The editing pipeline is the headline change. Users can draw arrows or lasso a region directly on the canvas, and the model reads that intent to make pixel-level changes inside the selection. Rough sketches and line drawings can be fed in as-is and rendered into finished designs. The Seed team wrote on its blog that the core breakthrough is "spatial positioning (grounding) and regional semantics" — the model knows which patch you circled and what role that patch plays in the composition. Prompts or bounding boxes can also decompose an image into layers that users then drag and rescale.
Multilingual generation is the second lever. Seedream 5.0 Pro natively handles Chinese, English, French, German, Russian, Japanese, Korean, Spanish and Arabic among more than ten commonly used languages, with third-party testers counting 14 languages rendered natively. Letterforms and typography follow local conventions instead of routing through an English prompt and back. That posture puts ByteDance head-on against Google's Nano Banana and OpenAI's GPT-Image, both of which lean English-first.
Pricing splits at resolution. Below 2.36 megapixels of output, the first input image is free, subsequent inputs cost 0.02 CNY each and outputs 0.3 CNY each. Above that, outputs jump to 0.6 CNY. Chinese outlet Cailianshe framed the release as ByteDance "benchmarking against Google's flagship product," and the cadence is aggressive — 5.0 Pro lands less than two months after Seedream 4.5.
One line of opinion: ByteDance did not chase GPT-Image on text comprehension. It stuffed Photoshop's marquee tool into a diffusion model.
via ByteDance Seed / ITBear / Cailianshe
The editing pipeline is the headline change. Users can draw arrows or lasso a region directly on the canvas, and the model reads that intent to make pixel-level changes inside the selection. Rough sketches and line drawings can be fed in as-is and rendered into finished designs. The Seed team wrote on its blog that the core breakthrough is "spatial positioning (grounding) and regional semantics" — the model knows which patch you circled and what role that patch plays in the composition. Prompts or bounding boxes can also decompose an image into layers that users then drag and rescale.
Multilingual generation is the second lever. Seedream 5.0 Pro natively handles Chinese, English, French, German, Russian, Japanese, Korean, Spanish and Arabic among more than ten commonly used languages, with third-party testers counting 14 languages rendered natively. Letterforms and typography follow local conventions instead of routing through an English prompt and back. That posture puts ByteDance head-on against Google's Nano Banana and OpenAI's GPT-Image, both of which lean English-first.
Pricing splits at resolution. Below 2.36 megapixels of output, the first input image is free, subsequent inputs cost 0.02 CNY each and outputs 0.3 CNY each. Above that, outputs jump to 0.6 CNY. Chinese outlet Cailianshe framed the release as ByteDance "benchmarking against Google's flagship product," and the cadence is aggressive — 5.0 Pro lands less than two months after Seedream 4.5.
One line of opinion: ByteDance did not chase GPT-Image on text comprehension. It stuffed Photoshop's marquee tool into a diffusion model.
via ByteDance Seed / ITBear / Cailianshe
