ByteDance V4.5
Unified generation-and-editing architecture with natural-language edits across up to 10 reference images.
Studio is in early access; availability resolves per project after sign-in.
- Endpoints
- 2
- Eligible
- 2
- Input schemas imported
- 2 / 2
- Delivery
- fal.ai
- Weights
- Not stated
- Developer
- Not stated in source
Overview
Seedream 4.5 edits images through natural-language instructions, processing up to 10 reference images simultaneously. Its unified architecture consolidates generation and editing, interpreting spatial references directly from prompts, for example replacing a product across figures, without layer masks or selection tools. Context-aware editing references multiple sources for composites, layout prototypes, and brand workflows.
Capabilities
- Natural-language instruction editing
- Up to 10 simultaneous reference images
- Mask-free spatial reference resolution
- Unified generation-plus-editing architecture
Best for
- Multi-image product composites
- Layout prototyping with text overlays
- Brand asset integration
Use cases
- Cross-figure product replacement
- E-commerce composite assembly
- Consistent-brand variations
Endpoints
2 eligible. Cataloged is not the same as executable: Studio resolves which endpoints a project can run.
| Endpoint | Task | Catalog status | Inputs |
|---|---|---|---|
fal-ai/bytedance/seedream/v4.5/edit | image editing | Eligible | 8 fields · requires prompt, image_urls |
fal-ai/bytedance/seedream/v4.5/text-to-image | text to image | Eligible | 7 fields · requires prompt |
Questions
How are edit targets specified?
In plain language with figure references; no masks needed.
How many sources per edit?
Up to 10 reference images at once.