Spaces:
Running on Zero
Running on Zero
File size: 3,113 Bytes
6936157 5f653cc 6936157 5f653cc 6936157 5f653cc 6936157 5f653cc 0eb3283 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | ---
title: StyleController
emoji: 🎨
colorFrom: purple
colorTo: indigo
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: Continuous image stylization with strength control
python_version: "3.12"
startup_duration_timeout: 1h
---
# StyleController
**Staying True to the Origin: Continuous Image Stylization with Smooth Transitions**
This Space demonstrates [StyleController](https://github.com/ReyChiaro/StyleController), a controllable
image stylization framework built on Qwen-Image-Edit-2509. Given a content image and a style reference,
it generates high-fidelity stylized images with explicit control over style strength via a slider.
## How it works
1. Provide a **content image** (the image you want stylized).
2. Provide a **style reference image** (the visual style to transfer).
3. Adjust the **style strength slider** (0 = no style, 1 = full stylization).
4. Click **Stylize** to generate the output.
The model combines a full-strength LoRA adapter with lightweight projector anchors learned at discrete
style strengths. B-spline interpolation between these anchors produces smooth, continuous transitions
while preserving content structure.
## Technical details
- **Base model:** [Qwen/Qwen-Image-Edit-2509](https://huggingface.co/Qwen/Qwen-Image-Edit-2509)
- **Adapter:** [ReyChiaro/StyleController](https://huggingface.co/ReyChiaro/StyleController)
- **LoRA rank:** 128, layers 0–59
- **Projector type:** low-rank linear
- **Interpolation:** B-spline (order 3, clamped knot vector)
- **Inference:** 16 steps, bf16, seed 42
## Citation
```bibtex
@misc{xu2026stayingtrueorigincontinuous,
title={Staying True to the Origin: Continuous Image Stylization with Smooth Transitions},
author={Rui Xu and Hanmo Zhang and Songhua Liu},
year={2026},
eprint={2608.08125},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2608.08125},
}
```
## License
StyleController source code is MIT-licensed. Adapter weights are Apache-2.0.
### Example image credits
**Content images** are from
[`linoyts/repo-to-space-example-inputs`](https://huggingface.co/datasets/linoyts/repo-to-space-example-inputs) (CC0-1.0).
**Style-reference images** each represent a distinct visual style category and
are drawn from public-domain / Creative-Commons sources:
- **Classical oil painting** — Claude Monet, *Wild Poppies Near Argenteuil* (1873),
Musée d'Orsay, via Google Arts & Culture / Wikimedia Commons. Public domain.
- **Watercolor** — John Singer Sargent, *Muddy Alligators* (1917), Worcester Art
Museum, via Wikimedia Commons. Public domain.
- **3D render / CGI** — *Classroom* Blender demo scene by Christophe Seux
(blender.org), via Wikimedia Commons. CC0-1.0.
- **Flat design / vector illustration** — *Vector Illustration of a Mountain
Landscape with Pine Trees* by Missaruthra, via Wikimedia Commons. CC0-1.0.
- **Anime / cel illustration** — flat anime-style illustration from
[`linoyts/repo-to-space-example-inputs`](https://huggingface.co/datasets/linoyts/repo-to-space-example-inputs) (CC0-1.0). |