Published: July 8, 2026
Image Model — generation
Seedream 5.0 Pro is BytePlus's image generation model, producing photorealistic images from text prompts with strong composition and detail.
It accepts up to 10 reference images to keep the same face, product, or object consistent across generations, and renders in-image text across 15 languages.
Its outputs are trusted inputs for the Seedance 2.x video models, so a Text-to-Image result can go straight into image-to-video.
Variants at a Glance
Type |
Modalities |
Max Res. |
Highlights |
|
|---|---|---|---|---|
| Variant | ||||
| 5.0 Pro (new) | Image generation | Text-to-Image, single-ref, multi-ref | 2K | Higher controllability and quality; up to 10 reference images; 15 native prompt languages |
| 5.0 Lite | Image generation | Text-to-Image, single/multi-ref, sequential | 4K | Lighter and faster; up to 14 reference images; adds sequential image generation; the 4K option today |
Seedream 5.0 Pro is the new model documented here; all Key Features and Technical Capabilities below describe Pro. Seedream 5.0 Lite is shown for comparison only.
Key Features
- Text-to-image: Generate images from a prompt with strong photorealism and composition
- Reference consistency: Keep the same face / identity, product, or object across generations using up to 10 reference images
- Photorealism & effects: Improved clarity, realism, atmosphere, and camera language for stronger visual storytelling
- In-image text & multilingual: Renders readable in-image text across 15 languages, with deeper local-culture accuracy for attire and patterns
- High-resolution output: 2K-class output across aspect ratios
Trust ecosystem with Seedance
- Trusted video inputs: Outputs are trusted by the Seedance family (2.5, 2.0, Fast, Mini), bypassing input moderation; Text-to-Image is trusted automatically, Image-to-Image after account identity verification (KYC)
Technical Capabilities
Details |
|
|---|---|
| Output resolution & tier | • 1.5K tier — ≤ 2.36M px, up to 1536×1536 • 2K tier — > 2.36M px, up to 2048×2048 • 16:9 — up to ~2.7K on the long edge (2752×1536) • Max total: 2048×2048 (4.19M px) |
| Aspect ratios | 1/16 to 16 — e.g., 1:1, 4:3, 16:9, 9:16 |
| Input images | Up to 10; single- or multi-image input (down from Lite's 14) |
| Input resolution | Width and height both > 14 px; ≤ 6000×6000 (36M px) per image |
| Input image size | Up to 30 MB per image |
| Input formats | JPEG, PNG, WebP, BMP, TIFF, GIF, HEIC, HEIF |
| Output formats | PNG / JPEG (default JPEG) |
| Prompt languages | 15 native: Arabic, English, Russian, Indonesian, Spanish, German, Turkish, Portuguese, Malay, Vietnamese, French, Japanese, Korean, Tagalog, Thai. Others work, but in-image text and local-culture accuracy are weaker. |
Limitations
- No 4K output — Pro ships 2K-class; use Lite for 4K
- Slower than Lite, as it targets higher controllability and quality
- No streaming / progressive output
- Text rendering is significantly improved but not perfect — occasional typos, especially small text and non-English scripts
- Caps at 10 input images, down from Lite's 14
Prompting Tips
- Include reference images to keep a face, product, or object consistent across generations
- Prompt in one of the 15 natively supported languages for the most reliable in-image text and local-culture accuracy
- To carry an image into video, generate with Seedream then pass it to a Seedance model; complete account identity verification (KYC) to make Image-to-Image outputs trusted too