| --- |
| title: Aegis Art Atelier |
| emoji: 🖼️ |
| colorFrom: blue |
| colorTo: yellow |
| sdk: gradio |
| sdk_version: "6.20.0" |
| app_file: app.py |
| pinned: false |
| short_description: Melkov, a fine-tuned Qwen2.5-VL-7B art expert agent |
| --- |
| |
| # Aegis Art Atelier - Melkov |
|
|
| Melkov is a vision-language art expert built on a QLoRA fine-tune of |
| Qwen2.5-VL-7B-Instruct, trained on a curated, style-balanced subset of the |
| OpenBrush art dataset. This Space serves the merged 16-bit checkpoint: |
| [beaunix/Aegis-Art-Atelier-Qwen2.5-VL-7B](https://huggingface.co/beaunix/Aegis-Art-Atelier-Qwen2.5-VL-7B). |
|
|
| Drop a painting into the chat box and Melkov will describe its subject, |
| style, composition, lighting, color palette, and mood. You can also ask |
| general art-history questions. |
|
|
| ## Notes |
|
|
| - This Space runs on ZeroGPU: the GPU is attached on demand per request, so |
| the first response may take longer while it attaches. |
| - Responses stream in progressively rather than appearing all at once. |
| - The model was fine-tuned for descriptive analysis of paintings; results on |
| photographs or non-painting imagery may be less reliable. |
|
|
| ## Project |
|
|
| Part of the Aegis portfolio. See the model card for training details, |
| dataset composition, and evaluation notes. |
