How to call 16 image and video models with one async Task API — and poll until the job is done

Tabtab is one API for 16 AI image and video models (11 image, 5 video), behind a single async Task API. You do not get an image or a video back in the same request. You create a task, get a task ID, and poll its status until it succeeds or fails. That pattern matters because video jobs take seconds, and a callback is not guaranteed — polling is the reliable fallback.

Docs: https://tabtabapi.com/docs

1. Create a task
   Base URL: https://api.tabtabapi.com
   Auth: Bearer key starting with sk_
   POST /api/task/create with model and args, for example model seedance-2.5 and a prompt for a short clip. The response returns a task ID. Keep it. request_id and callback_url are optional.

2. Poll the status
   GET /api/task/status?task_id= with the same Bearer auth. Poll until the status is succeeded or failed. Do not assume a single poll is enough, and do not rely only on a callback. Use a loop with a delay and a timeout. When it succeeds, fetch the result file from the returned URL. Result files expire, so download them promptly.

3. Local files
   Reference inputs are URLs. For a local file, upload it first via POST /api/task/materials. Limit: 50 MiB per file, 20 uploads per minute per key, default retention 1 day.

4. What it costs — the five models to start with
   Pay-as-you-go, top-ups from $2, special group 0.80x. Prices below are the default-group prices checked on 2026-10-31; the pricing page notes settlement follows the successful upstream price, so recheck before you budget a large run.

5. Seedance 2.5: $0.20/s at 720p — a 10-second 720p clip is $2.00

6. Kling 4.0: $0.09/s at 720p — a 10-second 720p clip is $0.90

7. GPT Image 2.5, per-image entry (gpt-image-2.5-r, not the per-token flare/sunburst entries): from $0.006/image at 1K — 20 images at 1K are $0.12

8. Seedream 5 Pro: $0.02/image

9. Nano Banana 2.1: $0.015/image

10. A checklist before you run a batch

11. Start with one task and one poll loop, not a batch.

12. Set a timeout and handle failed explicitly.

13. Download result files before they expire.

14. Price the batch first: seconds for video, images for image models, at the resolution you actually need.

No free tier is claimed here, and no uptime or latency figure is claimed — the selling point is one API, documented polling, and prices up front.