I built an App that judges your face to play music (So you don't have to scroll Spotify) 🎧

We've all been there. You open Spotify or YouTube, stare at your 47 different playlists, and spend 15 minutes trying to figure out if your current existential dread requires Lo-Fi Beats or Heavy Metal.

I decided that making decisions is overrated. Why not let a neural network look at your face and decide how you feel instead?

Say hello to Viso Vibe: https://viso-vibe.streamlit.app/

On the outside, it looks like a sleek, high-end, premium audio interface. On the inside? It’s a beautifully stitched-together monster of Python, custom JavaScript hacks, and ONNX models running inside a container.

Here is exactly how this absolute unit of an app works:

The Streamlit Catfish Frontend: Streamlit is amazing for data apps, but it usually looks like... a data app. I didn't want that. So, I brutally injected raw HTML, CSS, and JS to completely murder the default styling. Custom #06060F dark background? Checked. Noise textures? Checked. Turning a standard file/camera input into a glowing, animated shutter button? Absolutely.

The Dual-Model Brains (ONNX): When you click that glowing shutter, your photo is converted into a NumPy array and fed into the pipeline. First, detection.onnx hunts down your face, draws a box, and crops it. Then, emotion.onnx uses a neural network (HSEmotionRecognizer) to analyze your facial muscles and calculate a probability score across 7 baseline emotions.

The Emotional Router: To avoid choice paralysis, I boiled human emotion down into 4 distinct musical lanes:

1.Happiness/Surprise ➔ Happy (Solar Palette ☀️)

2.Sadness ➔ Sad (Cobalt Palette 🌊)

3.Anger/Disgust/Fear ➔ Angry (Crimson Palette 🔥)

4.Neutral ➔ Calm (Sage Palette 🌿)

The Passive-Aggressive Feedback: Before the track even starts, Google Text-to-Speech (gTTS) generates a custom voice track saying, "You seem [emotion]. Let me play something for you." It base64 encodes it into an invisible HTML tag to voice-line you the second the page updates.

The Ninja Media Player: Streamlit's audio players are static and boring. So, I used streamlit.components.v1.html to inject a massive custom UI card. It renders an HTML canvas orbital visualizer with floating particles and reactive audio bars. To get the actual music, a completely hidden div loads the official YouTube IFrame API. The app triggers a random index between 0 and 50 within my curated mood playlists, and the custom player buttons pipe invisible commands like player.playVideo() right to the hidden YouTube element. No YouTube player layout visible, just vibes.

🔥 Try it out & Roast/Review it!
It is live right now: 👉 https://viso-vibe.streamlit.app/

Go ahead, try to fake a smile to get a happy playlist, or give it your worst Monday morning stare to trigger the Angry (Tamil Mass Beats) queue.

I need your help with two things:

Playlist Suggestions: What absolute bangers, deep-cut YouTube mixes, or hidden lo-fi tracks am I missing? Tell me what needs to go into the Happy, Calm, Sad, or Angry databases!

Code Roast: Tell me what you think of using hidden IFrames and raw JS to bypass UI limitations.

Drop your thoughts and playlist recommendations in the comments! 👇