Felliks

Felliks / DoomVLM

Public

AI plays Doom — pit Vision Language Models against demons and each other. Solo scenarios, deathmatch arena, 1-4 agents with any OpenAI-compatible API

13
0
100% credibility
Found Mar 16, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Jupyter Notebook
AI Summary

DoomVLM is a Jupyter notebook project that allows vision language models to play Doom by analyzing game screenshots and issuing simple shoot or move commands through tool calls.

How It Works

1
🔍 Discover DoomVLM

You find a fun GitHub project where smart AI characters learn to play the classic shooter game Doom by seeing the screen and making moves.

2
🤖 Set up AI helpers

Download a simple app and pick lightweight AI brains to run on your computer, so they can think and decide during the game.

3
🎮 Prepare the game world

Install the game tools needed to run Doom scenarios right on your machine, creating arenas full of demons and challenges.

4
📱 Launch the control panel

Open the interactive guidebook, set up your AI players with names and styles, and pick a solo challenge or team battle mode.

5
Choose your game style
🎯
Solo Adventure

One AI takes on monsters alone in mazes or shootouts to build skills.

🏟️
Battle Arena

Multiple AIs clash head-to-head in deathmatches, racing to survive and score.

6
▶️ Start the action

Hit play and watch live as your AIs spot enemies, shoot across the screen grid, dodge fire, and fight in real time with scoreboards updating.

🏆 Enjoy the victories

Celebrate your AIs' wins, review highlight videos with health and kill stats, and share the epic moments from their Doom adventures.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is DoomVLM?

DoomVLM is a Jupyter Notebook that lets vision language models play classic Doom via ViZDoom, turning game screenshots into simple shoot or move decisions through OpenAI-compatible APIs. You get real-time gameplay where AI agents tackle 11 solo scenarios like defending corridors or gathering health, or battle in deathmatch arenas with 1-4 agents against each other or bots. It's the latest in ai plays doom experiments, like mouse brain plays doom or jesus plays doom, but with any VLM backend like LM Studio.

Why is it gaining traction?

It stands out by simplifying VLM testing in embodied environments—no RL training needed, just plug in your API endpoint for instant deathmatch arena PvP or benchmarks. Developers dig the live visualization, GIF/MP4 recordings with stats, and easy multi-agent setups where faster models dominate in realtime mode. Hooks like pitting models against demons echo viral github plays and doom played on anything vibes, making it a quick way to benchmark tool-calling on dynamic tasks.

Who should use this?

AI researchers evaluating VLMs for vision-action loops, like robotics sims or agentic workflows. RL engineers comparing model performance without custom envs, or indie devs prototyping dsp plays doom-style demos. Perfect for teams with OpenAI-compatible APIs testing 1-4 agents in competitive scenarios.

Verdict

Fun prototype for VLM benchmarking (13 stars, 1.0% credibility score), with solid docs and quick LM Studio setup, but early-stage—expect tweaks for production use. Try it if you want arena battles now; skip if seeking battle-tested RL baselines.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.