In September 2026, ByteDance released Seedream 5.0 Flash, an AI model for generating and editing images. It is a fast, low-cost version in the Seedream 5.0 series, and its key feature is that it creates images faster and more cheaply while having the same features as the higher-end Seedream 5.0 Pro. On Volcano Ark, the developer service of Volcano Engine, ByteDance's cloud for China, it is listed as a new model for September.
The announcement
According to Volcano Engine's documentation, Seedream 5.0 Flash (model ID doubao-seedream-5-0-flash-260915) is a model for high-precision image generation, and its "features are exactly the same as Seedream 5.0 Pro" (except for the fast mode described below). On top of that, the documentation explains that because it creates images faster and costs less, it is suited to work where you want to keep wait times and costs down.
The Seedream 5.0 series currently has three models: Pro for high image quality, the fast, low-cost Flash and the existing Lite. It is expected to be used in situations where speed and volume matter more than the highest quality of each image, such as ad images and product images that go through many revisions, or creating large numbers of design drafts.
What it can do
The main features available in Seedream 5.0 Flash are as follows.
- Generating images from text, and generating images based on one or more reference images
- "Interactive editing," which fixes only part of an image by specifying the location with coordinates, bounding boxes or arrows
- "Layer splitting," which splits one image into one background image and up to 16 layers
- A feature that, when a single image with a transparent background is input, edits and generates while keeping the background transparent
- A feature that draws text in a specified language within an image
In layer splitting, each layer is output as a PNG image with a transparent background, and the stacking order, position, name and description of each layer are returned along with it. This makes tasks such as swapping out only the text, logo or people in a finished image easier, much like in a design tool.
In addition to Chinese and English, prompts (instruction text) are supported in many languages listed in the documentation, including Japanese, Korean, French, German, Spanish, Thai, Vietnamese and Indonesian.
Number of reference images and resolution
Up to 10 reference images can be passed per request. That is fewer than Lite's maximum of 14 and the same as Pro. Output resolution can be chosen from three levels, 1K, 1.5K and 2K, with 2K as the default. For example, a 16:9 landscape image at 2K is 2816×1584 pixels. Output at 4K is not supported (Lite supports 4K).
In addition, Pro lets you choose a "fast" mode that simplifies prompt optimization to generate faster, but Flash does not support this mode because it is a model that prioritizes speed to begin with.
Pricing and where it is available
On Volcano Engine, the price is 0.12 yuan per output image. Images input as references are free. For regular image generation, Pro costs 0.30 yuan per image (1.5K or below) or 0.60 yuan (above 1.5K), so Flash is less than half that. However, Pro's price for splitting images into layers is 0.15 yuan (1.5K or below), so the difference from Flash is smaller.
For regions outside China, it is offered under the name Dola Seedream 5.0 flash on ModelArk, the AI development service of BytePlus, ByteDance's cloud for businesses (in the Singapore region; there is no guidance on whether it can be used from Japan). Pricing on BytePlus could not be confirmed in the official documentation. According to ComfyUI Wiki, a change adding Seedream 5.0 Flash to the feature for calling ByteDance models was merged into the image generation tool ComfyUI on September 24.
As for apps, according to IT Home, Doubao Work, ByteDance's AI for work, added support for Seedream 5.0 Flash as a new image model on October 9. No guidance from ByteDance could be confirmed on whether it can be used in Jimeng (Dreamina outside China), an image and video creation service, or in the video editing app CapCut. There is also no guidance on availability for Japan.
CoAI's take
Seedream 5.0 Flash is a model that keeps nearly the same features as the higher-end version while cutting the price of regular image generation to less than half. In particular, location-based partial editing and image layer splitting fit workflows in which generated images are not used as is but revised and finished afterward. In image generation AI, price cuts and stronger editing features are advancing at the same time, as seen in Google's Nano Banana 2.1 and Tencent's Hy Image 3.5 preview. Developers should check whether resolutions up to 2K and 10 reference images are enough, and then choose between Pro and Flash accordingly.
