A new project on GitHub, ReelMimic, promises to do something many video makers have wanted: take a video you admire and produce a fresh one in the same visual style. According to its GitHub repository, the tool accepts a reference clip and a one-line brief, then uses a team of AI agents built on Claude Code or Codex to plan, build and review the result. The project is published under the account edenfunf and is released under the MIT licence for its code.
How the workflow runs
The process starts with a reference. Users can supply a local file, a phone recording or a YouTube link. ReelMimic first breaks the reference into its technical parts: editing rhythm, shot count and length, beats per minute, transitions, colour palette, framing and camera movement.
Next, the tool produces a plan for the user to check. That plan includes a storyboard, the characters, the assets needed and a few style frames. Users can chat alongside the plan, change settings or add material before they approve it and start production.
Once production begins, the work is divided among several agents. The repository says up to six agents can work on different parts of the video at once. Each finished shot is then passed to a different agent for review, so no agent checks its own work. When a problem is found, the shot goes back for correction, and every fix must come with before and after screenshots that the reviewer examines.
What the tool does not copy
ReelMimic says it learns how the reference was made rather than reusing it. The project states that it does not carry over the original footage, characters or assets. Characters are created as original designs unless the user supplies their own, in which case the tool follows those designs. The repository also notes that anything the tool finds online is logged with its source, author and licence, and that unclear licences are flagged.
Styles, requirements and limits
The current release is limited to two-dimensional animation and offers seven drawing engines. These include vector and motion graphics built on HyperFrames, hand-painted watercolour, crayon picture-book, pixel art, paper cut-out stop-motion, whiteboard doodle and anime cel. The repository describes the five newer engines as having seen less real-world use than the first two. If a reference does not match a known style, the tool uses the nearest engine and writes a proposal for a new style.
Production takes time. The project says a 30 to 60 second video usually needs between one and 3.5 hours after the plan is approved, depending on length and look. Watercolour and crayon are the slowest, because each frame is painted.
Setup requires Node.js 22.18 or later, Python 3.10 or later, FFmpeg, Chrome and either Claude Code or Codex CLI, logged in. The repository lists the interface language options as Traditional Chinese, English and Simplified Chinese. Because all work runs through the user's own Claude Code or Codex account, it counts toward that account's usage limits. If a limit is reached, the job pauses and can resume later.
- Testing has mostly been done on Windows, with macOS and Linux described as likely to work but less tested.
- The tool produces animation only and is not designed for live-action footage of real people.
- Posting to social accounts is optional and requires a separate command through Upload-Post; nothing is published automatically.
Why it matters for creators
Tools that analyse pacing and framing could lower the cost of producing short animated content, particularly for small teams and independent creators. The design choice to split production across agents and require independent review is also notable, since it tries to address the common problem of AI tools delivering a finished result with no check on quality.
At the same time, the project is early. The repository's own roadmap lists plans for more characters and styles, and a way to save newly discovered styles so later references can match them directly. Users will also need to weigh the time required, the usage cost on their AI plan and the licensing questions that come with reference material.
ReelMimic is not a finished product, but it offers a clear example of how multi-agent workflows are being applied to creative production. Its progress will depend on how well the review process holds up across a wider range of styles and real projects.
Related: Reduce Claude Code Token Costs With optimAIzr CLI · Anthropic's OSS Scanner Automates Security Checks for Open Source
Comments (0)