Inspiration

Marketers needed UGC-style ads fast, but hiring creators, shooting, and editing slowed campaigns. We wanted a link-to-ad workflow that keeps quality high without the production overhead.

What it does

Takes any product URL, scrapes key info, auto-generates positioning and storyboard, lets you pick an AI avatar (100+ options) and language (50+), and delivers a ready-to-run UGC ad.

How we built it

  • Frontend: React + Vite + Tailwind.
  • Firecrawl scrapes images and descriptions based on the product URL
  • Tavily to perform market research and analyze current market trends
  • LLM model like Gemini to drafts UGC script and a visual prompt.
  • Audio: ElevenLabs turns the script into voiceover.
  • Video: Nano Banana generates the avatar image; Fal.ai (Kling model) convert image to video, optionally lip-synced with the audio.
  • Finish: Merge the voiceover and video (e.g., with FFmpeg/dubbing API) to output the UGC ad.

Challenges we ran into

  • Product pages are inconsistent: missing images, and anti-scrape website
  • Lip-sync and audio/video alignment are sensitive; mismatched timing or low-quality assets degraded perceived quality.
  • Quality of video and audio
  • Integration of API into the application

Accomplishments that we're proud of

  • Built the full prototype in under 24 hours
  • Making video and product details
  • Ability to comprehend new concepts easily

What we learned

  • Make it exist first, refine later
  • Having tasks distributed for each team members

What's next for UGC-AI

  • Avatar cloning
  • More avatar voices/tones and brand voice memory.
  • Built in video editor
  • Dashboard to show performance of each video generated on social media
  • Compliance & safety: consent tracking for avatars/voices, content safety checks, and export watermarks or audit logs when required.

Built With

Share this project:

Updates