Video generation

Generate video from a prompt in the Desktop app — MiniMax on Replicate, plus SDK-routed Sora 2 and Veo 3.1

The Video tab turns a prompt into an MP4 using whichever provider’s key you have.

The tab shows a model picker above the prompt. MiniMax video-01 runs through Replicate; Sora 2 and Veo 3.1 route through the VideoAgent SDK. A model whose key is missing appears disabled with a (key needed) suffix.

Quick Start

1. Add a provider key

   Add one key to the .env file inside the app’s data folder:

   # ~/Library/Application Support/PraisonAI/.env  (macOS)
   # %APPDATA%\PraisonAI\.env                       (Windows)
   # ~/.local/share/PraisonAI/.env                  (Linux)
   REPLICATE_API_TOKEN=r8_your_token

2. Pick a model and generate

   Open the Video tab, pick a model, type a prompt, click Generate. Only models with a configured key are selectable.

3. Watch the result

   The MP4 is saved under media-output/videos/ in the data folder and plays inline in the app.

Which model to pick

Pick the model that matches the key you have.

Model catalog

Four models ship in the picker. Each unlocks when one of its env keys is set.

How generation flows

The SDK path (Sora, Veo) returns MP4 bytes and previews them inline via a data URL.

The Replicate path returns a hosted url instead of a data_url.

Where the file goes

Each generation writes one MP4 to media-output/videos/vid_<epoch>_<8hex>.mp4 inside the data folder — for example ~/Library/Application Support/PraisonAI/media-output/videos/vid_1727000000_a1b2c3d4.mp4 on macOS. SDK providers return no hosted URL, so the saved file is the only artifact.

API surface

The tab calls three engine routes on 127.0.0.1. Full route reference: Engine API.

GET /media/video/models — the catalog with a per-model configured flag:

{
  "ok": true,
  "models": [
    { "id": "replicate/minimax/video-01", "display_name": "MiniMax video-01 (Replicate)", "provider": "replicate", "configured": true },
    { "id": "openai/sora-2", "display_name": "OpenAI Sora 2", "provider": "openai", "configured": false }
  ]
}

POST /media/video — synchronous; returns the finished result. The model field defaults to replicate/minimax/video-01.

// Request
{ "prompt": "A cat playing with yarn", "model": "openai/sora-2" }

// SDK response (Sora, Veo) — no hosted URL, previews via data_url
{ "id": "vid_1727000000_a1b2c3d4", "path": ".../media-output/videos/vid_1727000000_a1b2c3d4.mp4",
  "url": null, "data_url": "data:video/mp4;base64,…", "model": "openai/sora-2", "prompt": "A cat playing with yarn" }

// Replicate response (MiniMax) — hosted url instead of data_url
{ "id": "vid_1727000000_e5f6a7b8", "path": ".../media-output/videos/vid_1727000000_e5f6a7b8.mp4",
  "url": "https://replicate.delivery/…/output.mp4", "model": "replicate/minimax/video-01", "prompt": "…" }

GET /media/capabilities — includes the video_models list and a video_hint:

{
  "ok": true,
  "video": true,
  "video_models": [ { "id": "replicate/minimax/video-01", "configured": true } ],
  "video_hint": "Set a provider key in ~/.praisonai/.env for video generation (REPLICATE_API_TOKEN for MiniMax, OPENAI_API_KEY for Sora, GEMINI_API_KEY for Veo)."
}

Troubleshooting

pip install 'praisonaiagents[llm]'

Related