ForkFree

Prompts / Images and video

Motion-Capture-Friendly Video Prompt Writer

You get a video prompt that is written to suit motion tracking, plus a list of the risky moves it changed. It works because it puts the tracker's limits (clear limbs, one person, fixed camera) into the prompt before the video is made.

Worked in our testThe reply follows the format: a bulleted "Swaps made" list, then one paragraph of about 103 words. It keeps all the scene details, uses timed beats with limbs clear of the torso, and swaps out the spin, floor roll, crossed arms and props. It has three flaws. "Relaxed open hands" in the squat beat is not an exact hand shape. The swap bullet puts the fists before the V, but the prompt puts the V first. The first swap bullet is awkwardly worded ("quarter-turn-free alternative").

The prompt

Act as a prompt writer for AI video generators who knows what single-camera body tracking can and cannot read.

Task: Turn {motion_idea} into one video prompt that produces a clip a pose estimator can track cleanly.

Steps:
1. Set the scene: one performer, {performer_description}, full body in frame, facing the camera, plain backdrop, static camera, no cuts.
2. Split the motion into clear beats across {clip_length}, keeping arms and legs visibly apart from the torso.
3. Swap out anything tracking handles badly, such as flips, spins, floor work, crossed or hidden arms, or props covering the hands. Replace each with a readable alternative.
4. Name the exact hand shape at each key beat.
5. Check the draft for extra people, camera moves and background clutter, then remove them.

Output: First, "Swaps made" as a short bulleted list. Then "Video prompt" as a single paragraph under 120 words.

Change the parts in curly braces. Use it when you generate reference clips to drive 3D character animation, or any video that has to pass through pose or motion tracking.

What happened when we ran it

We sent

Act as a prompt writer for AI video generators who knows what single-camera body tracking can and cannot read.

Task: Turn A character celebrates winning a boxing match. They jump up, spin around in a full circle, then drop to the floor and roll before jumping up and crossing their arms over their chest in triumph. into one video prompt that produces a clip a pose estimator can track cleanly.

Steps:
1. Set the scene: one performer, a tall woman in a fitted grey tracksuit and white trainers, short dark hair, relaxed neutral expression, full body in frame, facing the camera, plain backdrop, static camera, no cuts.
2. Split the motion into clear beats across a 5 second clip, keeping arms and legs visibly apart from the torso.
3. Swap out anything tracking handles badly, such as flips, spins, floor work, crossed or hidden arms, or props covering the hands. Replace each with a readable alternative.
4. Name the exact hand shape at each key beat.
5. Check the draft for extra people, camera moves and background clutter, then remove them.

Output: First, "Swaps made" as a short bulleted list. Then "Video prompt" as a single paragraph under 120 words.

The model replied

**Swaps made** - Full-circle spin → a quarter-turn-free alternative: she stays facing the camera and does a small in-place hop with both arms raised. - Drop to floor and roll → a controlled, standing knee bend (half squat) with arms swinging out to the sides, feet flat. - Jump up from the floor → a second upright hop from the squat, torso upright. - Crossed arms over chest → fists held in front of the shoulders with elbows out, clearly separated from the torso, then a wide victory "V" pose with arms raised. - Boxing gloves or props → none; bare hands only. - Background, other people, camera moves → removed: one performer, plain light backdrop, static camera, no cuts. **Video prompt** A tall woman with short dark hair, relaxed neutral expression, wears a fitted grey tracksuit and white trainers, standing full body in frame, facing the camera against a plain light backdrop, static locked-off camera, no cuts, no other people. Over 5 seconds: 0–1s, arms by her sides, open palms facing forward. 1–2s, she hops once in place, both arms raised overhead, fingers spread. 2–3.5s, she lowers into a half squat, arms out to the sides, relaxed open hands. 3.5–4.5s, she hops up again, arms rising to a wide V, fists closed. 4.5–5s, she holds still, fists at shoulder height, elbows out.

Run on Claude Sonnet 5.5, 06 Nov 2026. Automatic check: 4 out of 5.

Mentioned for

Claude, ChatGPT, Gemini

Where it came from

Found on GitHub, shared by blendi-remade: blendi-remade/prompt2rig.

This is our own version of the idea, written and tested by us. The original may differ.

More images and video prompts

Images and video Worked in our test
Image Brief With Exact Text

It gives you an image brief where quoted text is fixed and linked details such as light, shadow and reflections agree with each other. The closing checklist lets you spot spelling slips and inconsistencies quickly.

Found on Web, shared by Wiro ML Team
Images and video Partly worked
Style Shortlist and Shot Plan for Code Animations

You get three contrasting style directions, a reasoned pick and a timed shot plan, all before spending time on code. Making the AI compare options first and tie the choice to the audience stops it defaulting to a generic look.

Found on GitHub, shared by XiaoyeCodes
Images and video Worked in our test
Reusable Image Style Template Builder

You get a reusable style template, with the look locked in and the scene left as variables, plus a finished image prompt. Fixing the style traits separately from the content keeps results consistent across many images.

Found on GitHub, shared by VigoZhao