VibeCheck

Inspiration

Creators constantly second-guess themselves before hitting publish — "does this actually communicate what I intended?" We wanted to give artists, designers, and content creators a way to stress-test their creative intent against a synthetic audience before they put their work out into the world. The gap between what a creator means and what an audience perceives is real, measurable, and fixable — we just needed to build the tool.

What it does

VibeCheck lets you upload a visual, describe the vibe you're going for, and instantly receive a structured feedback report: a 0–100 alignment score, perceived vs. intended vibe breakdown, reactions from six distinct audience personas (Gen Z trend chaser, brand marketer, indie artist, skeptical critic, casual viewer, community mod), perception insights on color/composition/mood, and concrete actionable suggestions to close the gap.

How we built it

Next.js 15 (App Router) for the full-stack framework Google Gemini multimodal AI for vision-based image analysis and persona simulation Prisma + PostgreSQL for project and analysis persistence NextAuth v5 for authentication Tailwind CSS with a custom dark design system for the UI OGL + WebGL for the interactive galaxy background animation

Challenges we ran into

Getting the AI to produce structured, consistent analysis output across wildly different image types ##was the hardest problem — prompt engineering the persona reactions to feel genuinely distinct (not ##just the same voice with a different label) required significant iteration. We also had to carefully design ##the scoring rubric so the alignment number felt meaningful rather than arbitrary.

Accomplishments that we're proud of

The alignment score genuinely surprises people — it's not just a number, it's a mirror. Seeing someone upload an image they were confident about and watch it score a 43 with a clear explanation of why is the moment the product clicks. We're also proud of how fast the pipeline is — full analysis in under 10 seconds.

What we learned

Synthetic audiences are a genuine design tool, not just a novelty. Framing AI feedback as persona reactions rather than generic critique makes it dramatically more actionable — people know how to respond to "the brand marketer reads this as corporate" in a way they don't respond to "the image lacks authenticity."

What's next for VibeCheck

Video and audio analysis pipelines (fields are already stubbed in the UI) Side-by-side revision comparison (re-run analysis and see the score delta) Shareable feedback reports for creative teams Custom persona creation so you can test against your specific audience

Built With

• bcryptjs
• glsl
• google-gemini-api
• hugging-face-inference-api
• next.js
• nextauth
• ogl
• postgresql
• prisma
• react
• sql
• tailwind-css
• typescript
• vercel
• zod