AI Models (Dashboard)

This page documents the internal image/video generation models used by the BlitzReels dashboard (AI Studio + Playground). These are powered by fal.ai.

If you're looking for the public REST API, start at Quickstart and the OpenAPI Spec.

Video Clip Routing Rule

When generating a single video clip:

• If a reference/source image is selected: we call an image-to-video model.
• If no image is selected: we call a text-to-video model.

This matters for Sora 2 because the image-to-video endpoint requires image_url, while the text-to-video endpoint does not.

Sora 2 (fal.ai)

We support Sora 2 Pro via fal.ai:

• Image to video: fal-ai/sora-2/image-to-video/pro
• Text to video: fal-ai/sora-2/text-to-video/pro

fal docs:

• https://fal.ai/models/fal-ai/sora-2/image-to-video/pro/llms.txt
• https://fal.ai/models/fal-ai/sora-2/text-to-video/pro/llms.txt

Notes:

• Supported durations (current app UI): 4s, 8s.
• Output resolution (current app config): 1080p.
• For privacy, we pass delete_video: false so the result can be stored/processed in the BlitzReels media library.

Other Supported Video Models (Dashboard)

The dashboard also supports other image-to-video and text-to-video models (Kling, MiniMax, Luma, Google Veo, Wan, xAI Grok).

The canonical list lives in code:

• src/lib/fal-constants.ts
• src/modules/playground/ui/video-generator-constants.ts