- 🧠 Mind-Brush Framework: A novel agentic paradigm that unifies Intent Analysis, Multi-modal Search, and Knowledge Reasoning into a seamless "Think-Research-Create" workflow for image generation.
- 📊 Mind-Bench: A specialized benchmark designed to evaluate generative models on dynamic external knowledge and complex logical deduction, exposing the reasoning gaps in current SOTA multimodal models.
- 🏆 Superior Performance:
- Elevates Qwen-Image baseline accuracy from 0.02 to 0.31 on Mind-Bench.
- Outperforms existing baselines on WISE (+25.8% WiScore) and RISEBench (+27.3% Accuracy).
| Search Case | Search & Reason Case |
|---|---|
case1.mp4 |
case2.mp4 |
1. Clone
git clone https://github.com/PicoTrex/Mind-Brush.git
cd Mind-Brush2. Install
conda create -n mindbrush python=3.12
conda activate mindbrush
pip install -r requirements.txt3. Configuration
Fill in the [required] fields in config.yaml (e.g., API keys, path settings).
Note
You can set your language in .chainlit/config.toml. In Default, the language is set to en-US. You can change it to Chinese by setting language = "zh-CN".
[We only support English and Chinese for now. If you want to support other languages, you can add the corresponding language settings in the locales folder.]
4. Launch
Run the application with the following command:
chainlit run app.py -wOnce started, access the dashboard at http://localhost:8000.