Seedream

作者：CivitaiOfficial

Seedream is ByteDance Seed's family of image generation models. It makes images from a text prompt and edits or combines up to seven of your own images, with strong prompt adherence and legible text inside the image, including Chinese characters.

Made by ByteDance Seed. All credit for the model goes to ByteDance. Civitai runs it through an API so you can generate on-site with no setup; there is nothing to download on this page.

Versions

• v5.5 (November 2026): ByteDance's newest release, with more natural lighting, finer materials and more refined skin textures. Renders at 2K.

• v5.0 Pro: renders at 2K.

• v5.0 Lite: a lighter 5.0 tier with a 2K/4K switch.

• v4.5: has a 2K/4K switch.

• v4.0 and v3.0: the earlier releases, rendering at 2K.

On Civitai

• Text to image and image editing: add up to seven images to edit, restyle or combine them.

• Aspect ratios 16:9, 4:3, 1:1, 3:4 and 9:16. Output size comes from the aspect ratio and, on v4.5 and v5.0 Lite, the 2K/4K switch; writing "4K" in the prompt does nothing.

• Seedream is API-only: there are no open weights, so LoRAs and other add-ons do not apply.

• Cost is per image and depends on the version; on v5.5, each reference image adds a little to the price.

Prompting

Write full sentences describing the subject, setting, lighting and style. For text in the image, put the exact words in quotes, such as a sign that says "OPEN LATE". There is no negative prompt; say what you want instead.

Links

• Seedream 5.5: Seedream 5.5 on ByteDance Seed

• Seedream 5.0 Pro: Introducing Seedream 5.0 Pro

• Seedream 3.0: Technical overview