talking-head-production — Claude Code skill: AI avatar videos
talking-head-production turns a still portrait plus a script into a talking-head clip: the skill walks the agent through portrait checks, voice choice and lipsync calls on the belt CLI.
no license 241 installs ★ 31 updated 2026-08-31
npx skills add 101-skills/superpowers --skill talking-head-production -a claude-codeWhat talking-head-production does
This is a production guide for avatar videos built on inference.sh models. It starts with the portrait rules: face in the middle of the frame, head and shoulders visible, eyes on the lens, nothing covering the face, at least 512x512 on the face area and preferably 1024x1024 or more. After that the agent gets ready-made belt app run calls to fill in with its own image and script.
P-Video-Avatar (pruna/p-video-avatar) is the one it points to first, with its own TTS covering 30 voices in 10 languages; the author puts it at 18x faster and 6x cheaper than the rest of the list. OmniHuman 1.5 handles two characters in one shot. OmniHuman 1.0 and PixVerse Lipsync do lipsync only, so the speech has to come from somewhere else, for instance falai/dia-tts.
Nothing runs before the belt CLI is installed and belt login goes through. Its allowed-tools entry is Bash(belt *) alone, so the agent will not wander off into other shell commands on its own. Takes past roughly 60 seconds are cut into segments and stitched back together with infsh/media-merger.
What it helps with
- turn one portrait and a script into a presenter clip
- check a headshot against the portrait rules before paying for a render
- pick an avatar model by speed and price
- join short segments into a longer video
- stage a two-person dialogue in a single frame
Example prompts for the agent
- Make a talking-head video from
portrait.jpgand read the script in the Zephyr voice. - Check this headshot against the skill's portrait requirements and tell me what to reshoot.
- Build a two-character dialogue with OmniHuman 1.5 and merge the segments.
How to install talking-head-production
1Skills CLI (Vercel)
npx skills add 101-skills/superpowers --skill talking-head-production -a claude-code2Manually: put the files in place
~/.claude/skills/talking-head-production/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/talking-head-production/
1Skills CLI (Vercel)
npx skills add 101-skills/superpowers --skill talking-head-production -a codex2Manually: put the files in place
.agents/skills/talking-head-production/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add 101-skills/superpowers --skill talking-head-production -a cursor2Manually: put the files in place
.cursor/skills/talking-head-production/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install 101-skills/superpowers guides/video/talking-head-production2Manually: put the files in place
.github/skills/talking-head-production/SKILL.md1Skills CLI (Vercel)
npx skills add 101-skills/superpowers --skill talking-head-production -a gemini-cli2Platform CLI
gemini skills install https://github.com/101-skills/superpowers --path guides/video/talking-head-productionManually: put the files in place
.devin/skills/talking-head-production/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add 101-skills/superpowers --skill talking-head-production -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about talking-head-production
How do I install talking-head-production in Claude Code?
Run npx skills add 101-skills/superpowers --skill talking-head-production -a claude-code. The skill then sits at ~/.claude/skills/talking-head-production/SKILL.md.
What does talking-head-production do?
It guides the agent through avatar video production: portrait validation, model choice, and the belt app run calls for voiceover and lipsync.
What does the skill need in order to work?
The belt CLI from inference.sh, signed in with belt login. Its allowed-tools entry is just Bash(belt *), so the agent stays inside that one command family.