InternVL
Open-source multimodal vision-language models (InternVL family) that understand images and video alongside tex
- Stars
- 10.2k
- Forks
- 792
- Licence
- MIT
- Language
- Python
- Last update
- 1.1 years ago
- Type
- AI model / agent
No commits in over a year — check for forks or alternatives before adopting.
What it is
InternVL is a family of open-source multimodal large language models that can take images (and some video) together with text and answer questions or hold a conversation about them. The project publishes model weights, training code and datasets, and the README positions it as an open alternative to commercial models like GPT-4o. The README shown is truncated, so details of specific model sizes and benchmarks are only partly visible here.
- Ask questions about the content of an image and get a text answer
- Run document, chart or diagram understanding tasks using the InternVL models
- Train or fine-tune a multimodal model using the released training scripts and datasets such as MMPR
- Evaluate a model on multimodal reasoning benchmarks
Great for
- Researchers comparing open multimodal models on vision-language benchmarks
- Developers who want to self-host a vision-language model instead of calling a paid API
- Teams fine-tuning or studying preference optimization (MPO) for multimodal models
- Students learning how multimodal LLMs are trained and evaluated
Not ideal for
- People who just want a finished consumer app; the README points to a hosted chat demo for that instead
- Users without GPU hardware or ML experience, since running the larger models requires substantial compute
- Teams needing guaranteed production support; the README does not describe a support commitment
How to get it
- 1
Clone and build from source
You'll need: Python · PyTorch with a CUDA-capable GPU (exact versions not visible in the truncated README)
Clone the repository
git clone https://github.com/OpenGVLab/InternVL.git - 2
Run it
- Read the Quick Start section of the README (linked as Quick Start with HuggingFace)
- Pick a model size from the HuggingFace collections linked in the README
- Load the model with the HuggingFace instructions in the Quick Start section
- Send an image and a text question to get a response
- 3
Get the most out of it
- Check the documentation site for FAQs and tutorials on preference optimization and domain adaptation
- Use the HuggingFace format if you want to work with the standard transformers library, as noted in the 3.5 release
- Try the hosted chat demo first to judge model quality before setting up local hardware
- Review the training shell scripts in the repo if you plan to fine-tune or reproduce results
Downloads
Strengths
- Model weights, training code and several datasets are published openly
- The project reports strong benchmark results among open-source models, per its news posts
- Multiple model sizes are provided, from small Mini-InternVL models up to very large ones
Trade-offs
- Large models need significant GPU resources, and the README does not give a complete hardware guide in the visible portion
- Benchmark claims are self-reported and come from the project's own news posts
- Documentation is split across the README, a docs site and external papers, so setup can take some reading
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| InternVL (this) | 10.2k | AI model / agent | advanced | 1.1 years ago | Researchers comparing open multimodal models on vision-language benchmarks |
| Swin-Transformer | 16.1k | AI model / agent | advanced | 2.3 years ago | Researchers reproducing results from the Swin Transformer paper |
| deep-research | 19.9k | AI model / agent | intermediate | 7 months ago | Developers who want a readable example of a deep research agent to learn from or extend |
| Auto-claude-code-research-in-sleep | 18.5k | AI model / agent | intermediate | 5 days ago | ML researchers who already use Claude Code or Codex and want a structured research workflow |
| generative-ai-for-beginners | 123k | Learning resource | beginner | 5 days ago | Developers who know some Python or TypeScript and want a guided introduction to generative AI |
| OpenHands | 92.5k | App | intermediate | 4 days ago | Developers who want coding agents running on a server while their laptop is off |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.