Video Generation API for Seedance, Kling & Wan | VizoAPI

What

VizoAPI is a single REST API and Credits wallet for calling multiple hosted video and image generation models, including Seedance 2.5, Seedance 2.0 Mini/Fast/Pro, Wan 3.0, Wan 3.0 Prime, Kling 4.0, MiniMax H3, Seedream 5.0 Pro and Seedream 5.0 Lite. It targets developers and product teams who want several model families behind one endpoint, one billing model and one documented integration pattern rather than separate provider contracts. Its positioning rests on published per-unit pricing, itemized per-task billing and no charge for failed tasks.

Features

• Unified generation endpoint: A single POST /v1/generations call with a model and input body returns a task ID immediately, letting one integration reach text-to-video, image-to-video, video-to-video, text-to-image and image-editing models.
• Async result handling: Results are retrieved by polling GET /v1/generations/{id} until closed_at is set, or delivered through signed webhooks using the Standard Webhooks specification, so long-running jobs fit standard server architectures.
• Credits billing with per-task itemization: Credits are held on acceptance and settled against actual usage, every charge maps to a specific task, failed or undelivered tasks release the hold in full, and Credits carry no expiry or subscription fee.
• Free request validation: POST /v1/estimates checks a request body without creating a task or holding Credits, returning the same validation errors as a create call.
• Agent-assisted setup: A published prompt for Claude Code, Codex or Cursor walks a coding agent through reading the project, the docs, model schemas and rate limits before adding server-side integration code.

Helpful Tips

• Compare prices at the specification you actually need: the advertised rates are each model's lowest standard rate (often 480p), while items such as Seedance 2.5 at 720p are quoted separately, so headline figures are not like-for-like across tiers.
• Treat create calls as non-idempotent. Automatic retries and double form submissions produce duplicate tasks and duplicate charges; after a timeout, look the task up by client_reference instead of re-sending.
• Model input schemas are not uniform: Kling and Wan take the prompt in input.prompt, while Seedance and MiniMax use a text part inside a content list, and unsupported fields are rejected with HTTP 422. Validate against each model's schema rather than assuming a shared body shape.
• Plan for the operational limits before launch: outputs are retained for 7 days only, there is no cancel call, unfinished tasks close as "unresolved" after the default 24-hour deadline, and default concurrency is 3 video tasks per organization, so a queue is required for volume.