Open-Source Skills That Write Your Meta Muse Video Prompts

Share

Several open-source projects on GitHub are built for making videos with Meta Muse, and each one covers a different stage. muse-video-skill turns an idea into a script, storyboard, and call instructions, muse-video-prompt-skill handles prompts, stitching, and frame checks for Muse's built-in video, muse-video-prompt asks questions to build a prompt, and muse-video-prompts is a library of 204 public prompts. The article compares their licenses and stated limits and includes a prompt you can give your AI agent to help you pick one.

Machine translation. Read the original

More and more people are making videos with Meta Muse, and a set of open-source projects built around it has appeared on GitHub. Some expand one idea into a full script and shot list. Others focus on writing detailed video prompts and stitching several clips together. One person has collected the Muse video prompts circulating online into a single library. We went through the four repos, sorted them by what they do, and listed the limits each one states in its own documentation.

What Each of the Four Projects Covers

Project What it covers Best for License
muse-video-skill Pre-production planning: scripts, shot lists, art direction, and call instructions for downstream models People who have an idea but can't write a shot list MIT
muse-video-prompt-skill Video generation inside Muse: writing prompts, stitching long videos, and checking frames People who want multi-shot films over 30 seconds MIT
muse-video-prompt Asks you one question at a time to expand an idea into a full prompt, and can reverse-engineer prompts from a reference video People who struggle to write a good video prompt No license file in the repo
muse-video-prompts 204 public Muse video prompts collected from the web People looking for ideas and prompts to adapt CC BY 4.0

The first three are skills. Once you install them in your AI agent's skill folder, you can call them just by chatting. The last one is a reference library, so you can simply browse it.

muse-video-skill: A "Virtual Film Crew"

It passes a video idea through six roles in turn: director, screenwriter, cinematographer, art department, VFX, and sound. In stages, it produces a script, a six-panel or nine-panel storyboard, a color palette, a prompt for each panel, and finally call instructions you can hand to a downstream video model. It also includes a cost estimate. The skill holds the shooting methods from 38 benchmark cases. You can say "use the style of a classic ad as reference," and it will pull the matching techniques.

Although the name includes Muse, it is a general-purpose planning tool. Its call instructions are currently built for Seedance 2.0, and the downstream list includes ComfyUI, Kling, and other tools. It does not generate images or video itself. The repo is still marked "in development." If you want to sort out what an ad needs before the shoot, read Advanced: Win Your AI Ad Before the First Shot alongside it.

muse-video-prompt-skill: Making Muse's Built-in Videos Longer

This is a hands-on workflow built specifically for Muse's built-in video generation. It writes prompts from a segmented template covering duration and aspect ratio, characters, setting, second-by-second action, camera movement, sound, dialogue, and things to avoid. Each single generation has a length cap, so it splits the film into beats, generates each segment in turn to continue from the one before, and uses ffmpeg to join them losslessly into a video over 30 seconds. After each segment is generated, it pulls frames automatically and notes where the video matches the prompt and where it drifts.

The author is candid in the README. Muse's built-in video has unstable face locking, so a character may turn into a different person across segments. In the author's region, Muse's built-in video generation is blocked (returning a 403 error), and as of November 5 it had not been restored. For now, the project only includes text examples.

muse-video-prompt: A Director That Asks You One Question at a Time

Just say what you want to shoot, and it fills in the details one question at a time: who is in the frame, how they're positioned, what they do, how the camera moves, and how the light and sound are arranged. At the end, you get a Chinese and an English version of the prompt, and the prompt goes through three rounds of checks. It can also work from a reference video. A script breaks the video into frames and builds a timestamped contact sheet, and then it reverse-engineers the prompt, separating what it saw from what it guessed. If you connect it to Muse4AI on your computer, it can also submit tasks automatically and check each segment.

Frame-by-frame analysis requires Python 3, FFmpeg, and Pillow installed on your computer. The repo currently has no license file. By convention, that means the author keeps all rights. Using it to learn is fine, but ask the author before modifying it or using it commercially.

muse-video-prompts: 204 Ready-Made Prompts

The collection gathers Muse video prompts that 68 creators publicly posted on X between mid-September and early November. Each entry credits its author and links to the original post. They are sorted by type: beat-synced multi-shot edits, boyfriend POV, selfie vlogs, dance, Chinese period style, animated CG, documentary landscapes, and more. Of these, 76 are full prompts, and the other 128 show excerpts only, with links to the original text.

The curator is an API service provider. The README states this relationship and notes that some prompts only work inside the Muse app and may not work through the API.

Have Your AI Agent Pick One and Install It

I'm making AI videos with Meta Muse and want to install an open-source skill to help. The three candidates are:
- LuoJiangYong/muse-video-skill (pre-production scripts, shot lists, and call instructions)
- masonmosen/muse-video-prompt-skill (prompts for Muse's built-in video, multi-segment stitching, and frame checks)
- dichenxiang/muse-video-prompt (step-by-step prompt interview, reverse-engineering from reference videos)

My situation: [e.g. I want to make a 30-second product short / I only want to turn one idea into a good prompt / I want to learn how others shoot their videos].

1. Read the README and SKILL.md of these three repos, and tell me in plain language which one fits my situation. Recommend one and explain why.
2. List what I need on my computer (for example ffmpeg or Python), and what their license allows me to do.
3. After I confirm, install it into your skill folder. Before installing, tell me where it will go.
4. Don't log in to Muse for me, and don't activate or buy any service.

To build a one-click prompt tool inside Muse, see Make AI video characters stop moving like puppets with a one-tap Muse pose library.

Notes

  • All of them are personal or small-team projects: Update frequency and stability are not guaranteed, so check the date of the latest commit before installing.
  • Muse's built-in video has regional restrictions: In some regions it is blocked outright. Whether it works for you depends on your own account.
  • The works in the prompt collection belong to their original authors: Studying them is fine. If you plan to reuse them commercially or republish them, CC BY 4.0 requires attribution, and if the original author has stated other terms, those apply.

References

Done with this one? Mark it to track your path.Sign in
Save it for laterSign in
Questions about this piece?Ask AI about this

Comments (0)

No comments yet. Start the conversation.

Sign in to join the discussion.Sign in