Just for funRoast Me Bangalore Style
Upload a Selfie, Get Roasted Like You're From Bangalore
A multimodal web app where Gemini roasts your photo in Bangalore English and a voice reads it back, word by word.
Rochester, N.Y. Anyone who grew up in Bangalore knows how a Bangalore roast sounds. I wanted to see if a model could pull it off. You drop in a photo, a Next.js route sends it to Gemini with a system prompt built around a slang guide I wrote by hand, and you get back a roast of two to four sentences. A second route turns it into speech with an Indian English voice from Azure, or ElevenLabs if you add your own key.
The fun part is the reveal. I decode the audio to get its exact length, then light up one word at a time so the text keeps pace with the voice, like karaoke. The text auto-sizes to fill the screen. To survive free-tier limits there's a fallback chain across 4 Gemini models: when one hits a rate limit, the next takes over.
The whole app took one day and three commits, and it's still live. Try it on your own photo.
How it works
Roast Me Bangalore Style, start to finish, in 4 steps.
- Step 1:
Upload
Drag in a photo
- Step 2:
Roast
Gemini with a Bangalore English system prompt
- Step 3:
Speak
Azure Neural TTS or your own ElevenLabs key
- Step 4:
Reveal
Words light up in time with the audio
Graphic: The Sahil Bachu