Generate videos using OpenAI Sora models with evaluation

This promptfoo example evaluates OpenAI's Sora video generation models, comparing sora-2 (standard quality, $0.10/second) against sora-2-pro (higher quality output, $0.30/second). Generated videos are stored in promptfoo's media storage at ~/.promptfoo/media/ and displayed in the web viewer with playback controls, thumbnail, and spritesheet preview support.

Use this example when you want to compare Sora model output quality and cost for a given prompt, or tune generation parameters like resolution and duration. Video generation is slow - expect 1 to 5 minutes per video - so this isn't suited to fast iteration; promptfoo caches results by request parameters (prompt, model, size, seconds) so re-running with the same inputs doesn't regenerate the video.

Configurable parameters include size (1280x720 or 720x1280, default 1280x720), seconds (4, 8, or 12, default 8), poll_interval_ms (default 10000) and max_poll_time_ms (default 600000) for controlling how generation status is polled, and download_thumbnail/download_spritesheet (both default true) for preview assets:

npx promptfoo@latest init --example openai-video
cd openai-video

export OPENAI_API_KEY=your_api_key_here

npx promptfoo@latest eval
npx promptfoo@latest view

Set OPENAI_API_KEY before running. The web viewer plays generated videos directly with metadata display, so comparing sora-2 and sora-2-pro outputs side by side doesn't require downloading files separately.

Teams evaluating OpenAI's Sora video models for quality, cost, and generation-time trade-offs before choosing one for a production video-generation workflow.