moondream
Tiny open-source vision-language model that runs on CPU, in ~2GB RAM — image captioning, VQA, OCR
Overview
moondream is a tiny yet capable open-source vision-language model (VLM) designed to run on consumer hardware without a GPU. Powered by a SigLIP vision encoder and Phi-2 language model, it handles image captioning, visual question answering, OCR, and object recognition in just ~2GB of RAM. Installation is one pip command, inference takes seconds on CPU, and it's fully Apache-2.0 licensed. Perfect for privacy-sensitive applications, on-device photo analysis, content moderation, and accessibility tools. The community Docker image (ilopezluna/moondream) provides an HTTP API for easy integration. With 9.8k+ GitHub stars, it's one of the most popular lightweight VLMs available.
Requirements
Min vCPU
1
Min RAM
2048 MB
Min Disk
10 GB
Rec vCPU
2
Rec RAM
4096 MB
Rec Disk
20 GB
Recommended VPS
Hostinger · KVM 2
2 vCPU · 8192 MB · 100 GB
Hostinger · KVM 2
2 vCPU · 8192 MB · 100 GB
Hostinger · KVM 4
4 vCPU · 16384 MB · 200 GB
Affiliate disclosure
Docker Compose
# Generated by Run This Ai — docker-compose.yml
services:
moondream:
image: ilopezluna/moondream:latest
restart: unless-stopped
ports:
- 8080:8080
volumes:
- ./data/moondream:/data
Related tools
Stable Diffusion WebUI
Generative image models in your browser
ComfyUI
The most powerful and modular diffusion model GUI with a graph/nodes interface for Stable Diffusion
whisper.cpp
High-performance C++ port of OpenAI Whisper for fast local speech recognition
Fooocus
AI image generator focusing on prompts and generating — a Midjourney-like experience offline
Coqui TTS
Open-source deep learning toolkit for text-to-speech, battle-tested in research and production
ChatTTS
High-quality conversational text-to-speech model optimized for natural daily dialogue