Seedream 5.0 API: ByteDance's Image Model via a Job-Based Endpoint (2026)
LivePair AI ·
Seedream 5.0 — ByteDance's image generation line — is a core lane in LivePair's media catalog, with Pro and Edit variants for the quality and modification passes. The access shape is async: submit a prompt, poll a job id, get a media URL. Prepaid per-image billing, no subscription, no GPU rental.
The Seedream lineup in the catalog
Three variants, three jobs: `seedream-5.0` for generation, `seedream-5.0-pro` for the quality ceiling, `seedream-5.0-edit` for image-to-image modification on an input you supply. The Pro tier costs more per image — the catalog endpoint returns the live rate for each.
Sibling coverage matters for comparison work: `qwen-image-3` and its Pro/Edit variants run in the same catalog with the same job lifecycle, so A/B testing image models is a string swap, not a new integration.
Generate — submit, poll, fetch
Step one: `POST https://livepairai.com/v1/agent/generate` with your `x-api-key` and a body of `{ model: 'seedream-5.0', prompt: '...' }`. The response is a `jobId` — media generation is async everywhere on this platform.
Step two: poll `GET /v1/agent/jobs/<jobId>` with the same key until `status` is `done`. The response carries a media URL — fetch it and save it, because renders auto-delete after 48 hours. The `livepair.py` client in our examples repo wraps this whole lifecycle in one call, with typed errors and a configurable timeout.
| Field | Value |
|---|---|
| Submit | POST /v1/agent/generate → { jobId } |
| Poll | GET /v1/agent/jobs/<id> → status: queued|running|done|error |
| Model ids | seedream-5.0 · seedream-5.0-pro · seedream-5.0-edit |
| Output TTL | 48h — download immediately |
| Billing | Per image, prepaid credits; failed jobs don't settle |
Where Seedream earns its rate
Seedream's profile leans toward photographic rendering and strong text-in-image fidelity — the lane where prompt-adherent product shots and poster work live. For stylized or illustration-heavy output, `qwen-image-3` is the comparison to run first; for edit passes on an existing asset, the dedicated Edit variant beats prompt-hacking a generation model.
Production pattern: generate cheap on the base tier for drafts and layout exploration, re-run the winners on Pro for the ship asset. Same key, same endpoint, different model string — the quality/price dial is runtime-configurable.
The rest of the media catalog
Image isn't the whole surface: `wan-3.0` and `wan-3.0-prime` handle video generation, `seedance-2.5-private` covers the members-only lane, and utility models (upscalers, background removal) chain after generation for a full pipeline. All of it runs through the same submit/poll contract — learn it once, the whole catalog opens.
Tooling shortcuts: `livepair` CLI (`npm i -g livepair`) wraps generate/models/balance in a terminal, the ComfyUI nodes expose the catalog inside workflows, and the MCP server lets agents call generation as a tool.
FAQ
- How do I call the Seedream API?
- POST to https://livepairai.com/v1/agent/generate with an lp_ key and { model: 'seedream-5.0', prompt: '...' }, then poll /v1/agent/jobs/<jobId> until done. The livepair.py client in our examples repo does the whole lifecycle in one call.
- What does Seedream 5.0 cost per image?
- Per-image billing from prepaid credits — check GET /v1/agent/models for the live rate. The base tier covers drafts; Pro is the quality ceiling; no subscription either way.
- Is there an edit/img2img variant?
- Yes — 'seedream-5.0-edit' takes a source image plus an edit prompt for image-to-image modification, same submit/poll contract.
- How long are generated files kept?
- 48 hours — the job response returns a media URL you should download immediately.