rea-video breaks a finished video down shot by shot, and Treg puts Seedance, Veo and Kling behind one pay-per-clip token
rea-video is a Claude Code and Codex tool that works out, shot by shot, whether a video was filmed, generated or built in code, and writes a remake plan for each shot. Treg, from the Superdesign team, now routes agents to Seedance, Veo, Wan, Hailuo and Kling video models with one token at the provider's price.
rea-video in one paragraph. JuneYaooo's Apache-2.0 tool went from zero to about 7,900 stars in two weeks. You tell your agent to install it and point it at a video file. It splits the video into shots and the audio into voice, music and effects, then gathers evidence: frame cadence, sensor noise, whether the whole image moves as one flat plane, AIGC tags in the metadata, plus checks of hands, text, physics, BPM and key. For each shot it decides whether it was live footage, AI video, an avatar, 3D, code-built motion graphics or a screen recording. Then it writes a plan: setups and lighting for live shots, a prompt and model choice for AI shots, and a Remotion approach for code shots. Captions come out with timings (SRT/ASS export), along with cut pacing and where each sound effect lands.
Two outputs. You get an HTML or PDF report to read, and a VIDEO.md package for the agent. VIDEO.md is an open format the author also published. It's layered from a few-hundred-token card down to one file per shot, so an agent reads only what a question needs instead of a breakdown that can run to hundreds of thousands of tokens. v0.1.0 (Nov 11) fixed that package format and separated sound effects into their own track.
Accuracy claims. The project's own eval puts the filmed/generated/coded call at about 98 to 99.6% on original files with tools alone (offline and free), dropping to about 89% on recompressed platform downloads unless you add a vision model (Gemini or Doubao) or let a vision-capable agent review shots. Shot-boundary recall is listed at 95% on nine hand-labelled clips. Those are the maintainer's numbers on their own test set.
Use it to learn, not to copy. The README's own FAQ says it's for studying structure, pacing and technique and then remaking with material you have rights to, without reusing anyone's footage, music, characters or likeness. That's the useful frame for a studio. Break down a reference ad to see how many setups and generations it really took, then budget your own version. A companion repo, capsule-cinema, automates AI-shot remakes, and the README carries a sponsor link to ChatCut.
Treg for video. Treg (superdesigndev, about 7,900 stars) bills itself as 'OpenRouter for agent tools': one base URL and token for thousands of paid endpoints, metered per call from a prepaid balance. On Nov 13 Jason Zhou launched its video page. Agents can call Seedance, Veo, Wan, Hailuo and Kling, plus image models and Fish Audio voices, through the same token. Treg says it charges what the provider reports with 0% markup and releases the hold if a job fails. Its own example lists a 14-second Seedance 2.5 480p take at about $0.103 per second through Treg against $0.22 per second on fal, and pitches the whole run as roughly 80% cheaper than Higgsfield. Those are Treg's comparisons using public list prices from Nov 11. The launch post passed 320 likes and 70k views by midday.
Caveats for Treg. 'Open source' needs an asterisk. The licence is Apache 2.0 with additional terms, including a restriction on running it as a hosted service, so read it before you self-host commercially. The launch demo starts by scraping viral TikTok posts as references and casting 'AI influencers'. If you try it, use your own or licensed references and original characters, and label AI content where platforms require it.
Signals: rea-video has creator and community posts plus a Trendshift badge, but no appearance in our GitHub Trending snapshots and no tracked curator pick. Treg's video launch is the company's own post. Neither is HOT by our rule.
Related tools on ReelStack
Skills & repos in our directory
- rea-video · Skill, Agentic video production
- Comfy Previs · Repo, Generation
- seriesmith · ComfyUI, Generation
- Awesome Short Drama Skills · Skill, Generation
More from the Lab
- Instagram's Edits for Mac has a built-in MCP server, and the community has wired Claude Code into all 274 editor tools
- ComfyUI v0.45.0 for video makers: RelayVSR video upscaling, H3-Image, Seedream 5.5 and Breeze TTS 2 nodes
- Scenario's agent skills pack is trending: images, video, audio and 3D from Claude Code, with a price check first
- Runway Dev lands in Claude Code, Codex and Cursor. Here's what the plugin actually does