IntroSVG-Qwen2.5-VL-7B
Learning from Rendering Feedback for Text-to-SVG Generation via an Introspective Generator–Critic Framework
Accepted by CVPR 2026 🎉
Model Summary
IntroSVG-Qwen2.5-VL-7B is an end-to-end vision-language model that generates high-quality SVG (Scalable Vector Graphics) code directly from natural language descriptions. The model is fine-tuned from Qwen2.5-VL-7B-Instruct through a multi-stage training pipeline that combines supervised fine-tuning (SFT), curriculum learning, chain-of-thought (CoT) reasoning, and direct preference optimization (DPO).
The defining feature of IntroSVG is its introspective generator–critic framework: a single unified model alternates between two roles — generator (producing SVG code) and critic (rendering and evaluating its own output) — enabling an iterative generate → evaluate → refine loop at inference time.
Model Details
| Property | Value |
|---|
| Base model | Qwen/Qwen2.5-VL-7B-Instruct |
| Parameters | ~7B |
| Architecture | Vision-Language Model (VLM) |
| Modalities (input) | Text prompts and rendered SVG images (during the critique stage) |
| Modality (output) | SVG source code |
| Training data | SVG-1M (custom corpus, ~1M samples) |
| Training paradigm | SFT → DPO with curriculum learning and CoT |
| License | Apache 2.0 |
Method Overview
The model is built through three core stages:
1. Data Construction
A mixed corpus is synthesized using an early-checkpoint model and a teacher VLM, comprising three subsets:
- Direct generation ($\mathcal{D}_G^{\text{direct}}$) — text-to-SVG pairs
- Correction ($\mathcal{D}_G^{\text{correction}}$) — flawed SVGs paired with refinements
- Critique ($\mathcal{D}_C$) — rendered SVGs paired with critique feedback
2. Supervised Fine-Tuning (SFT)
A unified VLM is trained on the mixed dataset, simultaneously acquiring:
- SVG generation capability
- SVG critique capability
3. Direct Preference Optimization (DPO)
A teacher VLM scores generated preference pairs, which are used to further optimize the generator policy $M_{\text{Policy}}$ via the DPO loss.
Introspective Inference Loop
At inference time, the same model performs a closed-loop introspective process:
- Generate an initial SVG from the prompt.
- Switch to the critic role: render the SVG and evaluate it.
- Assign a quality score based on the critique.
- If unsatisfactory, use the critique to guide the next round of correction.
This loop allows the model to refine its outputs iteratively without any external evaluator.
Intended Use
Primary use cases
- Text-to-SVG generation for icons, simple illustrations, logos, diagrams, and UI elements
- Programmatic vector graphics design as a creative co-pilot
- Research on vision-language reasoning, code generation, and self-refinement methods
Out-of-scope use
- The model is not intended for generating photorealistic raster images.
- It is not optimized for generating extremely complex artwork or production-ready brand assets without human review.
- It should not be used to produce misleading, infringing, or otherwise harmful imagery.
How to Use
Installation
1# 1. Clone the repository
2git clone https://github.com/gitcat-404/IntroSVG.git
3cd IntroSVG
4
5# 2. Create environment
6conda create -n introsvg python=3.10 -y
7conda activate introsvg
8
9# 3. System dependency for cairosvg (Linux)
10sudo apt update
11sudo apt install libcairo2 libcairo2-dev
12
13# 4. Python dependencies
14pip install torch==2.5.1+cu124 torchvision==0.20.0+cu124 \
15 --index-url https://download.pytorch.org/whl/cu124
16pip install -r requirements.txt
Download model weights
1pip install huggingface_hub
2hf download gitcat404/IntroSVG-Qwen2.5-VL-7B \
3 --local-dir Models/IntroSVG-Qwen2.5-VL-7B
Inference (recommended: lmdeploy server)
We recommend serving the model with
lmdeploy for accelerated inference. Example with 4 GPUs:
1CUDA_VISIBLE_DEVICES=0,1,2,3 lmdeploy serve api_server \
2 "Models/IntroSVG-Qwen2.5-VL-7B" \
3 --tp 4 \
4 --server-port 23333
Then run the introspective inference loop on a CSV of prompts:
1python inference_loop.py \
2 --MODEL_NAME Models/IntroSVG-Qwen2.5-VL-7B \
3 --CSV_FILE example/test.csv \
4 --OUTPUT_DIR your_output_folder
An example prompt file is provided at example/test.csv in the GitHub repository — each row contains one text prompt for SVG generation.
Quick start with transformers
1from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
2
3model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
4 "gitcat404/IntroSVG-Qwen2.5-VL-7B",
5 torch_dtype="auto",
6 device_map="auto",
7)
8processor = AutoProcessor.from_pretrained("gitcat404/IntroSVG-Qwen2.5-VL-7B")
9
10prompt = "A minimalist red apple with a green leaf."
11messages = [{"role": "user", "content": [{"type": "text", "text": prompt}]}]
12
13text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
14inputs = processor(text=[text], return_tensors="pt").to(model.device)
15
16output_ids = model.generate(**inputs, max_new_tokens=2048)
17svg_code = processor.batch_decode(
18 output_ids[:, inputs.input_ids.shape[1]:],
19 skip_special_tokens=True,
20)[0]
21print(svg_code)
💡 To unlock the full introspective refinement loop (generate → render → critique → correct), please use inference_loop.py from the official repository — it handles SVG rendering and feeds the rendered image back to the model in its critic role.
Training
All experiments were conducted on
8 × NVIDIA A800 GPUs, using the
LLaMA-Factory training pipeline.
Required artifacts:
Place the data under LLaMA-Factory/data/ and launch training with:
For DPO and the full multi-stage recipe, please refer to the scripts and configs in the
official repository.
Limitations
- Visual complexity ceiling. Highly intricate scenes, dense compositions, or fine-grained textures remain difficult to express in SVG and may produce simplified outputs.
- Text rendering inside SVGs can be imperfect (font substitution, kerning artifacts).
- Latency. The introspective loop trades inference time for quality; single-pass generation is faster but less polished.
- Language coverage. Training prompts are predominantly English; performance on other languages may degrade.
- Rendering dependency. The critic stage requires a working
cairosvg / Cairo installation to rasterize intermediate SVGs.
Citation
If you find IntroSVG useful in your research, please cite our paper:
1@article{wang2026introsvg,
2 title = {IntroSVG: Learning from Rendering Feedback for Text-to-SVG Generation
3 via an Introspective Generator-Critic Framework},
4 author = {Wang, Feiyu and Yang, Jiayuan and Zhao, Zhiyuan and Zhang, Da and
5 Li, Bingyu and Liu, Peng and Gao, Junyu},
6 journal = {arXiv preprint arXiv:2603.09312},
7 year = {2026}
8}
Acknowledgements
This work builds on the excellent open-source ecosystem around:
- Qwen2.5-VL — base vision-language model
- LLaMA-Factory — training framework
- lmdeploy — inference acceleration
- cairosvg — SVG rasterization
License
This model is released under the
Apache 2.0 license. Please ensure your use of the model also complies with the license terms of the underlying
Qwen2.5-VL-7B-Instruct base model.
