# Seedream 4.5

Seedream 4.5 upgrades the Seedream line with stronger editing consistency. Subjects, lighting, and color tone hold across edits, and stylization and prompt adherence step up against Seedream 4.0.

- **Model ID:** `bytedance/seedream-4.5`
- **Type:** image
- **Providers:** bytedance
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** $0.04/image
- **Canonical page:** https://vercel.com/ai-gateway/models/seedream-4.5

## Supported parameters

- **Parameters:** `seed`
- **Operations:** `text-to-image`, `image-to-image`
- **Input modalities:** `text`, `image`
- **Output modalities:** `image`
- **Aspect ratios:** Not reported
- **Resolutions:** Not reported
- **Durations:** Not reported
- **Frame rates:** Not reported
- **Maximum outputs per request:** Not reported
- **Maximum input images:** 14
- **Input image formats:** Not reported
- **Maximum input image size:** 10 MB

## Example

```ts
import { experimental_generateImage as generateImage } from 'ai';

const result = await generateImage({
  model: 'bytedance/seedream-4.5',
  prompt: 'A red balloon on a wooden table.'
});
```

## About

Seedream 4.5 was released December 3, 2025 as ByteDance's latest in-house image generation model. It builds directly on Seedream 4.0 and focuses the architectural updates on editing fidelity rather than new input modalities.

The headline change is editing consistency. When you ask Seedream 4.5 to alter a portrait, swap a background, or refine a small text element, subject details, lighting, and color tone carry over more reliably than in Seedream 4.0. That matters for product photography touch-ups, brand-controlled marketing assets, and any workflow where the source image already passed approval and only specific elements should change.

Stylization and prompt adherence also improve. Seedream 4.5 interprets descriptive style instructions more precisely, which reduces the number of regeneration passes needed to land on the intended look. ByteDance also calls out gains in portrait refinement, small-text rendering, multi-image composition, and visual aesthetics relative to Seedream 4.0.

Seedream 4.5 is a dedicated image generation model. Call it through the AI SDK's `generateImage` function with the model set to `bytedance/seedream-4.5`. AI Gateway handles authentication and provider routing.

## What to consider

If your pipeline runs many edit passes on the same source image, the per-pass cost compounds. Seedream 4.5 is listed at $0.04 per image; weigh that against Seedream 4.0 if your workload doesn't need the editing improvements. See N/A for resolution and quality tiers.

## When to use

### Best For

- Edit-heavy workflows where the source image is locked and only specific elements should change without disturbing the rest
- Portrait refinement and small-text rendering where Seedream 4.0 left visible artifacts
- Stylized creative direction that depends on accurate adherence to descriptive prompt language
- Multi-image composition where subject identity must hold across a coherent set

### Consider Alternatives When

- Editing fidelity isn't a priority and cost per image is the binding constraint. Seedream 4.0 sits at a lower per-image rate
- Web-connected retrieval and real-time context matter for the output. Seedream 5.0 Lite adds that capability
- The workflow needs video rather than still images. Seedance 2.0 or Seedance 2.0 Fast cover that

## Best for

- **Edit-Heavy Workflows:** Source-image edits that preserve subject detail, lighting, and color tone across passes
- **Portrait and Small-Text Rendering:** Refinement work where Seedream 4.0 left visible artifacts
- **Stylized Creative Direction:** Stronger prompt adherence on descriptive style instructions reduces regeneration passes
- **Multi-Image Composition:** Subject identity holds across a coherent set of related images

## Consider alternatives

- **Cost-Sensitive Generation:** Seedream 4.0 sits at a lower per-image rate when editing fidelity isn't the priority
- **Web-Grounded Output:** Seedream 5.0 Lite adds web-connected retrieval for real-time context
- **Video Generation:** Seedance 2.0 or Seedance 2.0 Fast cover text-to-video and image-to-video pipelines

## Frequently asked questions

### How does Seedream 4.5 differ from Seedream 4.0?

Seedream 4.5 improves editing consistency, stylization, and prompt adherence over Seedream 4.0. Subjects, lighting, and color tone hold better across edits, and the model also gains in portrait refinement, small-text rendering, multi-image composition, and visual aesthetics.

### Does Seedream 4.5 support multi-image inputs?

Yes. Seedream 4.5 retains multi-image composition capability and improves subject consistency across the generated set compared with Seedream 4.0.

### When is Seedream 4.5 the right choice over Seedream 4.0?

Pick Seedream 4.5 when editing fidelity matters: portrait touch-ups, small-text changes, brand-controlled refinements, or any pipeline where the source image is approved and only specific elements should change.

### How do I call Seedream 4.5 through AI Gateway?

Set the model to `bytedance/seedream-4.5` and call it through the AI SDK's `generateImage` function. AI Gateway routes the request through `bytedance` using your API key or OIDC token.

### What does Seedream 4.5 cost per image?

Seedream 4.5 is listed at $0.04 per image. See N/A for the full pricing breakdown.

### Does Vercel AI Gateway support Zero Data Retention for Seedream 4.5?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/seedream-4.5)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://docs.byteplus.com/en/docs)
- [Provider pricing](https://docs.byteplus.com/en/docs/ModelArk/1544106)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.