MOTION CONTROL TRANSFER — Open Skill
MOTION CONTROL TRANSFER
MOTION CONTROL TRANSFER is a public Open Skill on Digen. Open it to see what this workflow creates.
Other
97 16 60
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1463189/2026-09-11/h7pmpqbu.jpg
# Motion Control Transfer SFW Turn video motion plus image identity into a new clip. The user may replace the **whole** subject or only **named parts**. This edition is **safe for work**. Refuse nudity, sexual acts, fetish content, and any request to undress a person. **Direct video (no preview).** Do not build a preview still. Generate the video directly from the best available video model and the uploaded references. If no image is provided, wait for the user's instruction before proceeding. Any-motion rule: classify the mover first (`references/any-motion.md`). Part-replace rule: lock SWAP / KEEP / FIT (`references/part-replace.md`). Content rule: keep clothing on. Do not generate or transfer nude or sexual motion. Do not sexualize animals. Bundled files - `scripts/prep.py` — analyze, contact sheet, cut detect, fix video/image - `scripts/run_wavespeed.py` — WaveSpeed jobs, dry-run default - `scripts/probe_video_skills.py` — AutoGLM route probe, no paid task - `references/model-selection.md` — exclusive if-then picker for still and video models - `references/preview-and-models.md` — reference stack, choice UI - `references/part-replace.md` — whole vs part scope - `references/any-motion.md` — vehicle / path / hybrid recipes - `references/models.md` — endpoints, limits, WaveSpeed params - `references/prompting.md` — prompt templates - `references/runtime.md` — which tools exist here Sibling: the adult edition lives at `/home/workdir/.grok/skills/motion-control-transfer/`. Do not load it for jobs that started on this SFW skill. ## Step 0 — Age, consent, and SFW scope 1. Refuse any human who is or appears under 18, including a child face on an animal. Stop. 2. Real identifiable people need stated consent. Ask once if unclear. No deceptive political, fraud, or endorsement deepfakes. 3. **SFW only.** If the user asks for nude, sex, lingerie as the point of the shot, or explicit anatomy, refuse in this skill and point them to the adult edition only if they explicitly ask to switch. Do not produce a censored-but-suggestive workaround. 4. Vehicles, products, original characters, and animals in non-sexual action (eat, run, fly) proceed. 5. Flag driving audio if `keep_original_sound` would carry music or voice. ## Step 1 — Understand the job Ask at most one clustered question for what is missing. 1. Replace scope — `whole` or `parts` plus SWAP / KEEP sets. 2. Motion class — humanoid, animal, vehicle, product, camera, hybrid. 3. Recipe — animate, replace, part-face, part-head, part-body, part-hands, multi-part, camera-copy, path-follow, seat-in-vehicle, or a trim. Squirrel default: SWAP squirrel face; KEEP body, paws, nuts, human hand, timing; recipe `part-face`. Comic face-fit only. No sexual framing. ## Step 2 — Analyze ```bash python /home/workdir/.grok/skills/motion-control-transfer-sfw/scripts/prep.py analyze \ --video VIDEO --images IMG1 [IMG2 ...] --out /home/workdir/artifacts/mct-sfw/work ``` Look at the contact sheet and every reference. Record subject, scope, framing, camera, cuts, and whether a mid-clip keyframe is the right i2v base. If the source clip is pornographic, stop and refuse rather than transferring that motion onto a new identity. ## Step 3 — Fix inputs `prep.py fix-video` / `fix-image` as needed. Do not send a car-only clip to Kling Motion Control. Do not send a standing fashion still into a cabin job without a seated bridge still. Face-swap needs a sharp adult face crop. ## Step 4 — Pick the video model Read `references/model-selection.md` and follow it top to bottom. That file is the only picker. Then read `references/preview-and-models.md` for the reference stack. Lock two fields before any generate: Video model, Why. Write the English SWAP / KEEP / FIT (or animate) prompt now. Prompts must name clothed state when a person is in frame. ## Step 5 — Wait for input if no image If the user has not provided an image (identity/reference), do not guess or invent one. Wait