Motionmaxxing Github: What It Is and How to Use It

The “motionmaxxing github” question comes up because two different things share a similar name: an open-source video-editing repo people clone, and a set of Claude Code skills you install into a coding assistant. If you searched “motionmaxxing github” expecting a single repository that turns a talking head clip into an animated short, the closest match in this ecosystem is OPEN VIDEO EDIT, an open-source set of Claude Code skills at github.com/kevinbadi/open-edits. This post explains what that repo contains, how to run it locally, and how to wire the output into a scheduler so the video actually gets published.

  raw talking-head clip
          |
          v
  +-----------------------+
  |  OPEN VIDEO EDIT      |
  |  Claude Code skills   |
  |  (Python + ffmpeg)    |
  +-----------------------+
          |
          v
  animated short (mp4)
          |
          v
  creatoros media:upload clip.mp4
          |
          v
  creatoros posts:create --platforms ...

What Is Motionmaxxing on GitHub?

Motionmaxxing is a style of editing, not a library. The idea is that a plain talking-head video gets treated with motion: animated captions that bounce in, logos that slide, faces that pop, GIF inserts, b-roll cards and counters that tick up while you speak. You are not adding motion for decoration. You are adding it because a static frame on a phone screen loses attention in the first second.

On GitHub, the practical implementation of that style in this stack is OPEN VIDEO EDIT. It is an open-source repo of Claude Code skills that take a raw clip and render an animated short. It renders locally with Python and ffmpeg on your own machine, so nothing about your footage leaves your laptop unless you send it somewhere yourself.

The other half of the ecosystem is the Social Agents repo at github.com/kevinbadi/social-agents, which is a different tool: an agent harness that runs your social accounts through Creator OS. That one posts and replies. It does not edit video. Keep the two separate in your head.

Why the Motionmaxxing Github Repo Ranks for AI Video Motion

Search results for motion-heavy AI video editing are messy. Most pages are landing pages for closed tools. A GitHub repo answers the query directly: you get files, skill definitions and a render command. People searching “motionmaxxing github” usually want one of three things.

• They want to read the skill definitions to see how the motion is actually built.
• They want to run it locally without uploading client footage to a third party.
• They want to fork it and change the caption style or the animation timing.

A repository satisfies all three. A hosted editor satisfies none of them. That is the whole reason the repo ranks: it is the only answer that lets you open the source and change the numbers.

What’s Inside the Motionmaxxing Github Repository

The repo ships a set of Claude Code skills. Each skill is a folder with instructions that the assistant reads before it touches your footage. The skills listed in the repo are:

• split-animated-talking-head: crops the speaker and animates them against a background layer.
• longform-animated-talking-head: the same treatment for longer videos.
• fireship-style-edit and fireship-style-short: the fast-cut, high-motion caption style.
• split-animated-short: a vertical short version.
• reverse-engineer: reads a reference video and describes how its motion was built so you can reproduce it.

Under the hood it renders with Python and ffmpeg. There is no cloud render farm and no queue. You run it, ffmpeg does the work, and you get an mp4 in your output folder. If you have ffmpeg installed and working, the heavy part is already done.

If you want a narrated walkthrough of the whole setup, watch the launch video at youtube.com/watch?v=-QBJH_PK3pY and the build videos on the KevBuildsApps YouTube channel.

How to Install and Run Motionmaxxing From GitHub

The workflow is clone, install, invoke a skill. Here is the shape of it.

git clone https://github.com/kevinbadi/open-edits
cd open-edits
# install Python and ffmpeg dependencies
# then open the folder in Claude Code

Once the folder is open in Claude Code, the skills are available to the assistant. You do not run a CLI binary with a flag. You talk to the assistant and name the skill. A first prompt looks like this:

Use the fireship-style-short skill on ./raw/ep12.mp4.
Vertical 9:16, keep my face on the left third,
captions word-by-word, logo top right,
add a counter at the "40 percent" line.
Output to ./out/ep12-short.mp4

Two things make this work well. First, you describe the motion in terms of what the viewer sees, not in terms of ffmpeg filters. Second, you name the output path so you know exactly where the file lands.

For a longform episode, swap the skill name to longform-animated-talking-head and drop the vertical framing note. For a clip where you liked someone else’s edit and want to copy the feel, run reverse-engineer on that file first, read the description it produces, then feed that description into the edit skill.

The render is local, so the first run is slow while ffmpeg works through the timeline. Subsequent runs with the same settings are faster because your inputs are already on disk.

Motionmaxxing Github vs a Hosted Scheduler for Video Output

These are two ends of the same pipeline, and the choice is about where you sit.

Pick the GitHub repo when you want to own the look, fork the timing curves, or keep client footage local. Use Creator OS when the edit is done and the file needs to go out. The repo produces the file. The hosted scheduler handles everything after the file exists.

Common Motionmaxxing Github Errors and How to Fix Them

Almost every failure traces back to the environment, not the skills.

• ffmpeg not found. The skills shell out to ffmpeg. If it is not on your PATH, the render dies at the first frame. Install it and confirm with a version check in a fresh terminal.
• Wrong aspect ratio on output. Vertical skills assume a 9:16 source. If your clip is 16:9 and the subject sits center, the crop eats their shoulders. Re-frame in the prompt before you retry.
• Captions drifting out of sync. Long takes with pauses confuse word-level timing. Cut the dead air first, then render.
• Long render times. A ten minute source at high resolution will crawl. Render the short first, confirm the look, then do the long version.
• Skills not appearing in Claude Code. You opened the wrong folder. The assistant only sees skills in the directory tree you loaded.

Fix the environment before you edit the skill files. Changing caption logic will not save a missing ffmpeg.

Wiring Motionmaxxing Into Creator OS for Automated Video Output

This is the part most video repos leave out. You rendered a great short. Now it needs to go out, on the right platforms, with the right cover, at the right time. Creator OS covers Instagram, TikTok, YouTube, X, LinkedIn, Facebook and Threads from one place, and it ships a CLI so the step after the render is a command instead of a drag-and-drop.

Upload the file once and get a media id you can reuse everywhere:

creatoros media:upload ./out/ep12-short.mp4

Check the caption before you schedule it, because platforms have different ceilings:

creatoros validate:post-length --text "Your caption here"

Then create the post. Point it at the platforms and pass the media id:

creatoros posts:create \
  --text "Episode 12 is live. The 40 percent rule." \
  --platforms instagram,tiktok,youtube \
  --media med_abc123 \
  --title "Episode 12" \
  --scheduledAt 2025-06-14T17:00:00Z --timezone America/Toronto

Creator OS applies each platform’s own rules when it publishes. A vertical video on YouTube becomes a Short. A custom Reel cover is set with --cover. If you want it to sit in drafts instead of going live, add --draft. Networks you named but have not connected come back in missing_platforms, and the connected ones still publish.

From there, an AI agent can drive the same steps. Creator OS ships a hosted MCP server at mcp.creatoros.ca/mcp with tools like create_post, upload_media_from_url and get_post_analytics. You can also install 14 Claude Code skills with npx @creatoros/cli@latest init, which include post-shortform and post-longform. Details are in the CLI docs and MCP docs.

Two safeguards worth knowing. API keys are pinned to one workspace, so an agent holding one workspace’s key cannot post to another set of accounts. And read-only MCP connections can only see read tools; the API refuses writes from them. That is the setup covered in how to give an AI agent safe access to your social accounts.

If you are running this for clients, give each client its own workspace and its own agent, which is the shape described in one agent per client for agencies. If you would rather start from someone else’s harness than build your own, the Social Agents repo boots with npm start creatoros social-agents and interviews you about your brand before it starts posting and replying. Read more about that approach in what to automate in autonomous social media posting.

If your edits also feed short-form channels, the caption and layout pass in the motionmaxxing agent skill walkthrough pairs well with the scheduler step above.

For anything not covered here, the docs live at creatoros.ca/docs.

Get started with Creator OS

The point of moving the render into a repo is that the file stops being the finish line. Once the short exists, the publish step should be a command, not a manual upload on four apps.

Creator OS starts at $19.99/month or $59.99/year on the Creator plan, which covers up to 8 connected accounts: 7 socials plus Skool. API keys, the MCP server, the CLI and the agent skills are included in every plan, so nothing in this post costs extra. Agency plans start at $49/month for 5 sets of socials, and the ads add-on is $19.99/month on any plan with no percentage of ad spend.

Start at www.creatoros.ca/sign-up, then read the CLI docs to wire it into your render script. If you want to see the video side of the pipeline in action, the walkthroughs are on the KevBuildsApps YouTube channel.