gitroomhq Agent-Media UGC Playbook
You're an agent (Claude, Cursor, custom) that needs to produce a finished UGC video on agent-media. There is exactly ONE tool — make_ugc (Agent-Media UGC Video). You never pick a sub-skill; make_ugc resolves identity and runs the whole pipeline internally.
One tool, three shapes
| You want | Give make_ugc | Result |
|---|---|---|
| A short clip | a one-line script (+ optional person/image/character) |
one clean talking-head clip |
| A full monologue | a long script (+ a real face via image or character) |
a seamless multi-take video, never trimmed |
| A narrated b-roll review | script + broll_url + a character/image |
the person narrates over your footage |
How to call it
A single call. The server picks the pipeline, the take count and the duration, runs it in one Temporal workflow, and returns a skill_run_id you poll.
POST https://api.agent-media.ai/v1/skills/make_ugc/run
Authorization: Bearer $AGENT_MEDIA_API_KEY
{
"script": "Okay this is wild, I tried the new flow and it actually works.",
"person": "a friendly young woman, soft daylight, candid framing",
"name": "Maya, 27"
}
Poll with GET /v1/skills/runs/<skill_run_id> until status is succeeded; final_output.video_url is your finished MP4.
See skills/make-ugc/SKILL.md for the full field manual.
Step D — publish it (optional)
Once you have a video_url, you can post it straight to the user's TikTok / Instagram / X with the publish-to-social skill — POST /v1/social/publish (CLI agent-media social publish, MCP social_publish). The user connects each network once via OAuth (/v1/social/connect). See skills/publish-to-social/SKILL.md.
Identity
person— describe them in words;image— a photo (https URL or base64);character— a savedchar_…id orcharacter_sheet_urlfrom list_characters; omit all three for a default person.- A full monologue or a b-roll review needs a real face — pass
imageorcharacter, not justperson. - Reuse a saved character: pass the same
characteron the next call — no re-generation.
Rules
- make_ugc paces and chunks the script for you — pass the FULL line or monologue and do NOT trim it. Length is inferred from the script.
- Any image/video URL you pass must be a public https URL (make_ugc re-hosts it onto R2 for you) — or a
character_sheet_url/char_…for a saved character. - Credits are deducted as each take runs. Portrait + character-sheet prep inside a video are FREE — you only pay for the video itself.
- Don't put "selfie" or "phone" in a description — say "talking to camera" instead. make_ugc already handles this internally.
Troubleshooting
INSUFFICIENT_CREDITS— user is out. Surface the agent-media.ai billing page link.make_ugc_needs_face— a full monologue or a b-roll review needsimageorcharacter(a real face), not justperson.- Video has subject holding a phone — they used a pose hint mentioning "selfie" or "phone". Strip it or set
pose: "". - Step stuck in
submitted— Temporal worker may be restarting. Wait 30s and re-poll; if still stuck after 5 min, the workflow is hung — contact agent-media support with theworkflow_id.
See also
- reference/auth.md — first-time setup
- reference/pacing.md — 2–4 words/sec rule with examples
- reference/realism-rubric.md — the 9 realism props baked into every prompt