Generate, remix, poll, and download Sora videos via the OpenAI Video API.
npx skills add https://github.com/Rabbit-Ivan/Ivan-skills --skill sora-rabbit-ivan
Please help me install this Agent Skill. Skill: sora Source: https://github.com/Rabbit-Ivan/Ivan-skills/tree/main/skills/sora Command: npx skills add https://github.com/Rabbit-Ivan/Ivan-skills --skill sora-rabbit-ivan
scripts/sora.py). - Structured prompt augmentation: Converts plain prompts into production-oriented specs (scene, camera, lighting, palette, dialogue) with CLI flags or JSONL batch fields. - Batch generation: Run many prompts concurrently from a JSONL file with retry logic for rate limits and transient errors. - Use Case: A marketer needs a 4-second product teaser and a variant. Generate the base clip with create-and-poll, then use remix with stated invariants to change only the color palette. ## Quick Start Ask the agent to generate a 4-second Sora video of your subject using the sora skill, then download the finished MP4 once the job completes.High-intent search queries and answers about installing and using this skill.
Set OPENAI_API_KEY, then run the bundled CLI with `uv run --with openai python sora.py create --prompt "..." --size 1280x720 --seconds 4`. Video creation is async, so poll the job until it completes before downloading the MP4.
Use the remix command with the completed video ID and a prompt describing only the change, for example `remix --id video_abc123 --prompt "Same shot, shift palette to teal/sand/rust"`. State invariants explicitly to prevent drift from the original shot.
Models are limited to sora-2 and sora-2-pro. Durations must be 4, 8, or 12 seconds passed as a string parameter. sora-2 supports 1280x720 and 720x1280; sora-2-pro adds 1024x1792 and 1792x1024.
Failures occur when the size is not supported by the chosen model or seconds is not one of "4", "8", or "12". Match the size to the model and pass seconds as a string value to avoid invalid_type errors.
No. Input reference images must be jpg, png, or webp matching the target size, and images containing human faces are rejected by the API. Content guardrails also prohibit real people, copyrighted characters, and copyrighted music.
Write a temporary JSONL file with one job per line (each requiring a prompt key), then run `create-batch --input prompts.jsonl --out-dir out --concurrency 3`. The CLI writes one JSON response per job and retries transient rate-limit errors automatically.
graysurfConvert, resize, crop, and optimize images using ImageMagick CLI.
rfdiosuaoCreate short video clips from URLs with transcription and captions.
adamhjoujExtract video frames by time or index using ffmpeg.
SheshiyerCraft AI video prompts with shot type, lighting, and style keywords for Veo, Seedance, Wan, Grok, Kling, Runway, Pika, and Sora.