ViralLens: Generating Non AI YouTube ideas from Inspirations

Inspiration

Most creators, from complete beginners to mid-size channels, can’t figure out how to replicate what the biggest YouTubers do. And every AI idea tool gives the same generic ideas that read like AI wrote them. We wanted a tool that learns from what actually works and builds ideas around you.

What it does

ViralLens is a Gemini agent. You describe what you make and something real about yourself, and pick your creator size. The agent:

• Finds outlier videos in your niche on its own using the YouTube Data API.
• Watches them with Gemini’s native video understanding, breaking down the hook, pacing, format, and -emotional trigger, with clickable timestamps.
• Extracts the shared viral pattern plus an inferred audience profile.
• Writes fresh ideas sized to your tier and your real story, with AI clichés banned.
• Scores each idea on “would this feel human?”, flags AI-sounding phrases, and ranks ideas so you know what to make.

How we built it

A Python FastAPI backend and a single-page frontend with Tailwind. Gemini handles function calling for the agent, video input from YouTube URLs, structured JSON outputs, and an LLM-as-judge authenticity pass. Prompts were prototyped in Google AI Studio. We used the YouTube Data API v3 on Google Cloud for search and video stats, with disk caching for reliability.

Challenges we ran into

We hit Gemini free-tier rate limits fast, so we moved to a Google Cloud project key and cached every analysis. Getting ideas that don’t sound AI-generated took several rounds of prompt iteration and a separate skeptical scoring pass.

Accomplishments that we’re proud of

It works end to end: one input goes to an agent that searches, watches, writes, and self-critiques. And the authenticity scores are honest, flagging its own weak ideas instead of hyping everything.

What we learned

Multimodal video understanding is a real advantage over transcript-only tools. Self-evaluation loops make LLM output much more trustworthy. And scoping tightly beats adding features.

What’s next for ViralLens

Draft feedback: upload your recorded draft, or film in a studio with gemini where it can walk through scenes and each day, and Gemini watches it and critiques it frame by frame against the viral pattern, then iterates the idea with you.

Built With

• gemini
• googleaistudio
• googleyoutube
• javascript
• python
• tailwind-css
• youtube-data-api