What if your feed knew you better than you know yourself?
What if it generated content faster than you could consume it?
What if you couldn’t take your eyes off the screen — and it was all made just for you?
This is a full run of the app — the story, video prompt, text-to-speech, and music were all AI-generated.
What you see was tailored to the user’s viewing patterns in real-time, to maximize retention.
The Real Doomscroll is an AI-powered TikTok clone with infinite, personalized, addictive content — generated in real time by machines that don’t sleep and don’t care about your mental health.
This is the result of a 2-day hackathon experiment by Obvious, the French art collective exploring the frontiers of artificial intelligence and its impact on culture.
It’s a proof of concept, not a product — a conversation starter to examine where things might be heading.
In a world where video generators can produce content faster than humans can watch,
and where large language models can guess what you want before you know it yourself…
What happens when TikTok becomes a generative, reactive, infinite loop?
- 🧠 LLM:
Qwen/Qwen3-4Bgenerates short TikTok-style stories, video prompts, and voiceover scripts. - 🎬 T2V:
LTX-Videorenders the video in seconds from the prompt. - 🗣️ TTS:
hexgrad/Kokoro-82Mcreates an uncanny synthetic voice. - 🎵 Music:
ACE-Stepcomposes eerie, emotionally tuned background tracks. - 📱 The result: A fresh AI-made TikTok in your feed.
- ⏱️ Your behavior — how long you watch vs skip — is fed back into the loop.
- 🔁 The AI adapts. The scroll deepens.
Because we’re artists.
And because we’re optimists.
We believe artificial intelligence can and should be used for the benefit of humanity — to create, to explore, to imagine.
But with every powerful tool comes the risk of misuse.
The Real Doomscroll is not a startup.
It’s not a product.
It’s an artistic experiment to explore the edges of what's coming.
We built it not out of fear — but to raise awareness.
To ask questions before others blindly build.
To show, not just speculate.
To let people feel what it’s like — and decide what we should do about it.
Ban it?
Regulate it?
Use it differently?Let’s talk.
- 🧠 LLM:
Qwen/Qwen3-4B— story + prompt generation - 🎬 Video:
LTX Video— fast text-to-video rendering - 🗣️ TTS:
hexgrad/Kokoro-82M— voiceover generation - 🎵 Music:
ACE-Step— AI-generated background sound - 📈 Loop: Simple heuristic tracking of watch time for engagement-based content generation
Tested on a RTX Pro 6000 Blackwell
Obvious is a French trio of artists and researchers using AI to push the boundaries of creativity.
We were the first to sell an AI-generated artwork at Christie’s New York in 2018.
Since then, our work has been exhibited in major museums worldwide, and in collaborations with Nike, Alpine, and the Opéra de Paris.
In 2023, we launched our own AI research lab with Sorbonne University, dedicated to inventing new tools for artistic creation.
In 2024, we visualized human imagination from MRI scans using AI — a project showcased in Paris and Seoul.
We believe AI is not the enemy.
It’s a mirror.
What we do with it is what matters.
The Real Doomscroll is one such reflection.
This repo is currently a conceptual project. Code & setup instructions coming after refactor.
We are Obvious
📸 Instagram
🐦 Twitter / X
📧 [email protected]
We turn algorithms into art — and sometimes, into warnings.
Feel free to open an issue, or open your mind.