mode-marketing-ua

Mode AI Creative Loop

An AI creative studio that turns a short brief into finished, on-brand vertical (9:16) UGC-style videos for Mode's apps. You pick a presenter and answer a few quick questions; it writes the script, generates the AI presenter footage and B-roll, then edits everything together with animated captions, motion graphics, music, sound effects, and the app's call-to-action — and hands you ready-to-post videos, usually in several variations.

Built on the WAT framework (Workflows, Agents, Tools): plain-language playbooks guide the AI, which runs reliable, repeatable tools under the hood.

What you get

• Finished 9:16 videos (1080×1920) — often multiple variations from a single run
• An AI presenter (persona) plus B-roll cutaways
• Animated captions, motion-graphic cards (app UI, social proof, count-ups, "FREE"), background music, and sound effects
• The correct app CTA / endscreen added automatically
• On-brand copy and visuals throughout (persona voice, brand language, required disclaimers)

How to use it — one command

Everything runs through the /create-ugc-video command inside Claude. You don't touch code.

1. Start it — run /create-ugc-video <persona> <app> (e.g. /create-ugc-video single-mom-maria mode-earn).
2. Answer a few quick questions — it asks:
   • which persona (the on-screen presenter),
   • video length (~30s / ~45s / ~60s),
   • whether you'll provide a script or Claude writes it, and
   • Autonomy: On or Off — the one setting that decides how hands-on you are (see below).
3. Approve the script — this is the key checkpoint.
   • If Claude writes it, it drafts a few script options; you review, tweak until you're happy, and pick which ones to produce. Nothing is generated until you approve.
   • If you provide the script, it's used as-is.
4. It produces the videos — generates the presenter footage + B-roll, then edits, captions, scores, and renders each variation, and appends the app's CTA endscreen.
5. You get finished videos back, ready to post.

The one choice that matters: Autonomy On vs Off

• Autonomy On → once you approve the script, it runs straight through to finished videos. Fastest and most hands-off.
• Autonomy Off → it pauses after each step (location, presenter shots, B-roll, video clips, final edit) so you can tweak or approve before it continues. Most control.

Either way, it never spends generation credits before you've approved the script.

---

For whoever sets it up (technical)

Everything below is only needed to install and run the project on a machine — not for day-to-day use of the /create-ugc-video command.

First-time setup (fresh clone)

On a new Windows or Mac, follow workflows/setup_environment.md — or just tell Claude "set up the environment." It runs tools/env_check.py, installs anything missing (each failure prints the exact per-OS install command), builds the Python venv from requirements.txt, and helps you create .env from .env.example.

python tools/env_check.py        # shows what's missing (any OS, no venv needed)

Prerequisites

Python 3.11, ffmpeg/ffprobe, Node/npx, git, yt-dlp (in the venv); optional: gh, Bun, uv. env_check.py verifies them all. If the venv is recreated, also recreate .venv/Lib/site-packages/sitecustomize.py (forces UTF-8 stdio so helper output doesn't crash on Windows — see setup workflow step 4).

API keys (.env)

Real keys live in .env (gitignored — never committed). Copy .env.example to .env and fill in the values: Higgsfield (AI footage), ElevenLabs (transcription), Gemini (analyzes scraped competitor videos during research — Claude writes the scripts), plus optional Apify (competitor scraping), Postiz, HeyGen, OpenRouter. Get the values from your team admin.

Assets

The repo ships the reusable assets (CTA, brand, screenshots, reviews, script-writing guide, SFX, music, props, fonts, personas). Heavy per-video AI clips and rendered edits are not in git — they live in the team's Google Shared Drive.

Project structure

mode-ai-creative-loop/
├── .agents/skills/     ← bundled skills (video-use, hyperframes, higgsfield, …)
├── tools/              ← Python pipeline scripts (the WAT "tools")
├── workflows/          ← plain-language SOPs (create_ugc_video is the spine)
├── assets/             ← per-app brand, personas, CTA, SFX, music, script guides
├── requirements.txt    ← Python dependencies for the venv
├── .venv/              ← project Python env (regenerable, gitignored)
└── .env                ← API keys (gitignored)