Instructions to use Matteoooo46/ReCoEdit-RL with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Matteoooo46/ReCoEdit-RL with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,3 +1,49 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
base_model: Qwen/Qwen-Image-Edit-2511
|
| 3 |
+
library_name: peft
|
| 4 |
license: apache-2.0
|
| 5 |
+
tags:
|
| 6 |
+
- lora
|
| 7 |
+
- diffusion
|
| 8 |
+
- image-editing
|
| 9 |
+
- product-consistency
|
| 10 |
+
- grpo
|
| 11 |
+
- reinforcement-learning
|
| 12 |
---
|
| 13 |
+
|
| 14 |
+
# ReCoEdit — RL LoRA Adapter (Epoch 24)
|
| 15 |
+
|
| 16 |
+
LoRA adapter for **Qwen-Image-Edit-2511**, trained with Flow-GRPO and a product consistency reward.
|
| 17 |
+
Plug this into `Qwen/Qwen-Image-Edit-2511` to get a model that better preserves product identity (color, shape, logo) during image editing.
|
| 18 |
+
|
| 19 |
+
- **Base model**: `Qwen/Qwen-Image-Edit-2511`
|
| 20 |
+
- **Training**: GRPO (Group Relative Policy Optimization) on e-commerce product editing data
|
| 21 |
+
- **Reward**: Qwen3-VL-30B-A3B-Instruct as VLM judge, 1-5 product consistency score
|
| 22 |
+
- **Result**: Reward improved from 0.022 → 0.480 (21.8× over SFT baseline)
|
| 23 |
+
- **LoRA rank / alpha**: 64 / 128
|
| 24 |
+
- **Checkpoint**: Epoch 24 (peak performance)
|
| 25 |
+
|
| 26 |
+
## Usage
|
| 27 |
+
|
| 28 |
+
```python
|
| 29 |
+
from peft import PeftModel
|
| 30 |
+
from huggingface_hub import snapshot_download
|
| 31 |
+
|
| 32 |
+
# Load base model (via DiffSynth-Studio)
|
| 33 |
+
# See https://github.com/Matteoooo46/ReCoEdit for full inference script
|
| 34 |
+
lora_path = snapshot_download("Matteoooo46/ReCoEdit-RL")
|
| 35 |
+
pipe.load_lora(pipe.dit, lora_path + "/adapter_model.safetensors")
|
| 36 |
+
```
|
| 37 |
+
|
| 38 |
+
See the [ReCoEdit repository](https://github.com/Matteoooo46/ReCoEdit) for the complete inference pipeline including APG guidance and prompt rewriter.
|
| 39 |
+
|
| 40 |
+
## Citation
|
| 41 |
+
|
| 42 |
+
```bibtex
|
| 43 |
+
@misc{recoedit2025,
|
| 44 |
+
title={ReCoEdit: Rewriter-Guided Consistency Alignment for Product Image Editing},
|
| 45 |
+
author={},
|
| 46 |
+
year={2025},
|
| 47 |
+
howpublished={\url{https://github.com/Matteoooo46/ReCoEdit}},
|
| 48 |
+
}
|
| 49 |
+
```
|