Skip to main content

​
AI Video Generation Integration

This skill builds a complete video workflow into your product: model selection, reference uploads, task submission, polling, error handling, durable storage, and playback. It supports text-to-video, image-to-video, multimodal references, native audio, and authorized real-person scenarios.

​
What It Can Do

  • Generate from text, one or more images, or mixed image/video/audio references.
  • Use Seedance 2.5 for synchronized audio, virtual presenters, and authorized real-person videos.
  • Select from enabled Seedance, MiniMax, Kling, and Wan models.
  • Create idempotent async tasks and show queued, processing, completed, and failed states.
  • Copy temporary provider output into project storage for durable playback and sharing.
When this capability is enabled, superun prepares the required cloud runtime and file storage and installs the calling credential in the project’s server environment. You do not need a separate provider API key.

​
Current Model Capabilities

Seedance 2.0 Mini supports at most 12 total assets, and audio cannot be its only reference. Seedance 2.5 supports at most 50 total assets and allows audio-only references. The live model selector and server admission rules are authoritative.

​
What to Prepare

​
Enable in superun

1

Click Use

Open AI Video Generation Integration in the Skill Library and click Use.
2

Define the workflow

Specify the models, inputs, duration, aspect ratio, audio behavior, and history requirements.
3

Verify the non-billable flow first

Check validation, loading, polling, failure messages, and storage behavior before a user explicitly creates a real paid task.

​
Required Safeguards

  • Video generation is asynchronous and can take from about 30 seconds to several minutes. Prevent duplicate submits and reuse the same idempotency key when retrying creation.
  • Browser-local paths and blob: URLs are not valid provider inputs. Upload local references to project storage and submit a server-reachable HTTPS URL.
  • Model switching must clear unsupported duration, resolution, audio, and reference inputs before submission.
  • Seedance 2.5 real-person requests accept exactly one HTTPS portrait and require explicit lawful-rights confirmation. The server performs the private asset import; the app must not save or expose provider asset IDs.
  • Provider result URLs are temporary. Copy videos that must persist into project storage.
  • Keep the video credential in server-side functions only—never in browser code, database rows, logs, or responses.
  • Every real create request consumes credits. Do not call generation merely to probe a model, dropdown, or integration.