Make AI video characters stop moving like puppets with a one-tap Muse pose library

What you'll get

Build your own pose prompt library inside Meta Muse. Open it to browse poses, tap one, and get a prompt you can paste straight into a video or image tool. You don't need to write code or know how to draw. You just send a few messages to Muse. Three ready-to-copy prompts are included.

Muse builds the tool and writes the prompts. You pick the poses and judge whether the results look right. It won't decide which motion fits your video.

Why poses deserve their own library

When you make AI character videos, motion is usually what gives the fake away. The face can look great and the outfit can be detailed, but once the person moves, they look like a puppet and viewers notice right away. The problem is often the prompt. It says only "a girl walking" and leaves out which way the body faces, which foot holds the weight, and what angle the camera uses.

Writing those details from scratch every time is tiring. So save your common poses in one library, with a ready-made description for each. When you need one, tap it, copy it, and paste it into your generator.

Before you start, keep three things in mind:

• Use a reliable pose source. This guide uses an open-source Chinese pose gallery on GitHub, ai-pose-gallery-zh. It has 1,439 poses, and the categories and descriptions are already organized.
• Don't copy the images. The repo says the images belong to their original authors and cannot be used commercially. So the tool only uses simple stick-figure sketches that Muse draws, plus a link back to the original repo.
• One pose, several prompt styles. Video tools and image tools read prompts differently, so write each pose separately for each tool.

Step 1: Create a Space in Muse

A Space is a feature Muse launched in November 2026. Think of it as a small app that Muse builds from your description. Once it's ready, it shows up in the Muse sidebar like its own tab.

Open the Muse app or the web version, and tell the chat box that you want to build a Space. You don't need to find any button. If you haven't set up Muse yet, first make sure you can chat with it normally.

Step 2: Tell Muse what to build, all at once

This is the most important part of this guide. It gives Muse the data source, the page layout, the prompt style, and the things it must not touch, all in one message. That way you won't have to go back and forth.

Here's why it's written this way:

• Only two categories. The original gallery includes adult-oriented categories. Limiting the scope keeps the tool clean and helps you avoid platform rule problems.
• Bracketed placeholders. The library only covers poses. People and scenes change every time, so placeholders work much better than fixed text.
• Check 5 first. Five poses are enough to see whether the stick figures are crooked and whether the descriptions are clear. That's easier than fixing 60 later.
• You can change the number [60]. It's a good idea to start small and add more once you're comfortable.

Step 3: Check the result against this list

Once Muse hands back the sample, don't just say "looks good." Go through the checklist below first. If a row is wrong, send that row's original text back to Muse and ask it to fix it.

The most common problem is vague prompts, such as one that only says "standing pose." If you see that, tell Muse:

The prompts are too vague. Please add three things to each pose prompt: which way the body faces (front, side, or back to the camera), which foot carries the weight, and where each hand is. After the change, show me 3 examples.

Step 4: Generate with your prompts

Pick a pose, copy the prompt for the tool you use, replace the bracketed parts with your character and scene, then paste it into your video or image tool.

A few habits make the results more reliable:

• Make one action per video. Going from standing to looking back is enough. If you add walking, looking back, and waving all at once, the motion gets messy.
• Make a character reference image first. Use the image version to create a look you like, then use that image for image-to-video. It gives you more control than text-to-video alone.
• Don't skimp on camera descriptions. Words like "medium shot, eye level, slow push-in" often help more than changing the character description.

Step 5: Have it add new poses every week

Once the library is built, you can have Muse update it on a schedule, so you don't have to keep track yourself:

Every Monday at 9 a.m., pick 10 poses that are not yet in the library from https://github.com/anhao/ai-pose-gallery-zh, using only the categories "Full body, single, standard" and "Upper body, single". Add them to the "Pose Prompt Library" in the same format. Do not change how the prompts are written. After you finish, tell me which poses you added.

Muse will repeat the time and task back to you to confirm. After that, it runs on schedule. If you no longer want it, just say "cancel the weekly pose update."

Notes

• Spaces just launched, and they roll out in different regions at different times. If you can't find it in your Muse app, wait a while and try again.
• Images in the pose gallery cannot be used commercially. The stick figures Muse draws are your own. Don't download the original images and put them into work you sell.
• Generation tools update quickly. If your prompt template starts giving weaker results after a while, ask Muse to rewrite it for the newer tool version.
• The stick figures are only a guide for choosing poses. The final motion in the generated video or image is what matters, so if it doesn't look right, pick a different pose.

References

• Pose gallery: anhao/ai-pose-gallery-zh (GitHub)
• Meta Help Center: How to manage reminders and scheduled tasks with Muse

Want Muse to read your email every day and send you a summary? See Have Meta Muse read your Gmail on a schedule and send a summary.