Seedream 5.0

Seedream 5.0 is ByteDance's image generation and editing family, not a video model - the directory entry it came from describes it as text-to-video, which is wrong. It ships in three sizes (Pro, Flash and Lite) and reaches users two ways: through the Volcengine Ark API, and through ByteDance's own apps. Pro arrived on 8 July 2026 after a February preview, and the headline additions were region-based editing (select an area, describe the change, everything else stays put) and layer decomposition, which splits a finished poster into separate product, shadow, background and text layers as transparent PNGs.

API pricing is per image and depends on output size. On the official Ark model page, an output at or below 2.61 megapixels costs 0.3 CNY per image, anything larger costs 0.6 CNY, the first reference image is free and each additional one is 0.02 CNY; the pixel threshold matters more than the model name, because a well-sized 16:9 image can stay in the cheaper tier. Third-party routing is close to that: the Flash variant is listed at $0.018 per image with roughly 15-20 second generations, and fal and Replicate list Pro at $0.045 for 1K output and $0.09 for 2K. There is no free tier at the API level - every request is billed.

Two costs that surprise people. Layer decomposition bills per output image, so one request for a transparent asset can come back as a dozen billed images. And watermarking is provider-dependent: on some routes the 'AI generated' mark is only removed if you explicitly pass watermark off, and streaming or multi-image generation returns an error on models that do not support them. Quality-wise the reviews are consistent: strong on Chinese text rendering and dense poster layouts, weaker on faces and fine interface detail. It suits e-commerce and design teams producing high volumes of Chinese-language product imagery, not portrait work.