Ovis-Image

Built upon Ovis-U1, Ovis-Image is a 7B text-to-image model specifically optimized for high-quality text rendering, designed to operate efficiently under stringent computational constraints.

🏆 Highlights

• Strong text rendering at a compact 7B scale: Ovis-Image is a 7B text-to-image model that delivers text rendering quality comparable to much larger 20B-class systems such as Qwen-Image and competitive with leading closed-source models like GPT4o in text-centric scenarios, while remaining small enough to run on widely accessible hardware.
• High fidelity on text-heavy, layout-sensitive prompts: The model excels on prompts that demand tight alignment between linguistic content and rendered typography (e.g., posters, banners, logos, UI mockups, infographics), producing legible, correctly spelled, and semantically consistent text across diverse fonts, sizes, and aspect ratios without compromising overall visual quality.
• Efficiency and deployability: With its 7B parameter budget and streamlined architecture, Ovis-Image fits on a single high-end GPU with moderate memory, supports low-latency interactive use, and scales to batch production serving, bringing near–frontier text rendering to applications where tens-of-billions–parameter models are impractical.

✨ Showcase

Here are some examples demonstrating the capabilities of Ovis-Image.

🚀 News

• [2025/12/19] 🔥 Ovis-Image has been merged into vllm-omni!
• [2025/12/7] 🔥 Ovis-Image has been merged into stable-diffusion.cpp!
• [2025/12/3] 🔥 Ovis-Image has been merged into diffusers!
• [2025/12/2] 🔥 Ovis-Image has been merged into ComfyUI!
• [2025/11/29] 🔥 Announcing Ovis-Image (Model)!

🛠️ Inference

Inference with Diffusers

First, install the diffusers library with support for Ovis-Image.

# pip install git+https://github.com/DoctorKey/diffusers.git@ovis-image
# pip install git+https://github.com/huggingface/diffusers
pip install diffusers>=0.36.0

Next, use the OvisImagePipeline to generate the image.

import torch
from diffusers import OvisImagePipeline

pipe = OvisImagePipeline.from_pretrained("AIDC-AI/Ovis-Image-7B", torch_dtype=torch.bfloat16)
pipe.to("cuda")
prompt = "A creative 3D artistic render where the text \"OVIS-IMAGE\" is written in a bold, expressive handwritten brush style using thick, wet oil paint. The paint is a mix of vibrant rainbow colors (red, blue, yellow) swirling together like toothpaste or impasto art. You can see the ridges of the brush bristles and the glossy, wet texture of the paint. The background is a clean artist's canvas. Dynamic lighting creates soft shadows behind the floating paint strokes. Colorful, expressive, tactile texture, 4k detail."
image = pipe(prompt, negative_prompt="", num_inference_steps=50, guidance_scale=5.0).images[0]
image.save("ovis_image.png")

Inference with Pytorch

Ovis-Image has been tested with Python 3.10, Torch 2.6.0, and Transformers 4.57.1. For a full list of package dependencies, please see requirements.txt.

git clone [email protected]:AIDC-AI/Ovis-Image.git
conda create -n ovis-image python=3.10 -y
conda activate ovis-image
cd Ovis-Image
pip install -r requirements.txt
pip install -e .

For text-to-image, please run

python ovis_image/test.py \
    --model_path AIDC-AI/Ovis-Image-7B/ovis_image.safetensors \
    --vae_path AIDC-AI/Ovis-Image-7B/ae.safetensors \
    --ovis_path AIDC-AI/Ovis-Image-7B/Ovis2.5-2B \
    --image_size 1024 \
    --denoising_steps 50 \
    --cfg_scale 5.0 \
    --prompt "A creative 3D artistic render where the text \"OVIS-IMAGE\" is written in a bold, expressive handwritten brush style using thick, wet oil paint. The paint is a mix of vibrant rainbow colors (red, blue, yellow) swirling together like toothpaste or impasto art. You can see the ridges of the brush bristles and the glossy, wet texture of the paint. The background is a clean artist's canvas. Dynamic lighting creates soft shadows behind the floating paint strokes. Colorful, expressive, tactile texture, 4k detail." \

Alternatively, you can try Ovis-Image directly in your browser on

📊 Performance

Evaluation of text rendering ability on CVTG-2K.

Model	#Params.	WA (2 regions)	WA (3 regions)	WA (4 regions)	WA (5 regions)	WA (average)	NED↑	CLIPScore↑
Seedream 3.0	-	0.6282	0.5962	0.6043	0.5610	0.5924	0.8537	0.7821
GPT4o	-	0.8779	0.8659	0.8731	0.8218	0.8569	0.9478	0.7982
SD3.5 Large	11B+8B	0.7293	0.6825	0.6574	0.5940	0.6548	0.8470	0.7797
RAG-Diffusion	11B+12B	0.4388	0.3316	0.2116	0.1910	0.2648	0.4498	0.7797
FLUX.1-dev	11B+12B	0.6089	0.5531	0.4661	0.4316	0.4965	0.6879	0.7401
TextCrafter	11B+12B	0.7628	0.7628	0.7406	0.6977	0.7370	0.8679	0.7868
Qwen-Image	7B+20B	0.8370	0.8364	0.8313	0.8158	0.8288	0.9116	0.8017
Ovis-Image	2B+7B	0.9248	0.9239	0.9180	0.9166	0.9200	0.9695	0.8368

Evaluation of text rendering ability on LongText-Bench.

Model	#Params.	LongText-Bench-EN	LongText-Bench-ZN
Kolors 2.0	-	0.258	0.329
GPT4o	-	0.956	0.619
Seedream 3.0	-	0.896	0.878
OmniGen2	3B+4B	0.561	0.059
Janus-Pro	7B	0.019	0.006
BLIP3-o	7B+1B	0.021	0.018
FLUX.1-dev	11B+12B	0.607	0.005
BAGEL	7B+7B	0.373	0.310
HiDream-I1-Full	11B+17B	0.543	0.024
Qwen-Image	7B+20B	0.943	0.946
Ovis-Image	2B+7B	0.922	0.964

Evaluation of text-to-image generation ability on DPG-Bench.

Model	#Params.	Global	Entity	Attribute	Relation	Other	Overall
Seedream 3.0	-	94.31	92.65	91.36	92.78	88.24	88.27
GPT4o	-	88.89	88.94	89.84	92.63	90.96	85.15
Ovis-U1	2B+1B	82.37	90.08	88.68	93.35	85.20	83.72
OmniGen2	3B+4B	88.81	88.83	90.18	89.37	90.27	83.57
Janus-Pro	7B	86.90	88.90	89.40	89.32	89.48	84.19
BAGEL	7B+7B	88.94	90.37	91.29	90.82	88.67	85.07
HiDream-I1-Full	11B+17B	76.44	90.22	89.48	93.74	91.83	85.89
UniWorld-V1	7B+12B	83.64	88.39	88.44	89.27	87.22	81.38
Qwen-Image	7B+20B	91.32	91.56	92.02	94.31	92.73	88.32
Ovis-Image	2B+7B	82.37	92.38	90.42	93.98	91.20	86.59

Evaluation of text-to-image generation ability on GenEval.

Model	#Params.	Single object	Two object	Counting	Colors	Position	Attribute binding	Overall
Seedream 3.0	-	0.99	0.96	0.91	0.93	0.47	0.80	0.84
GPT4o	-	0.99	0.92	0.85	0.92	0.75	0.61	0.84
Ovis-U1	2B+1B	0.98	0.98	0.90	0.92	0.79	0.75	0.89
OmniGen2	3B+4B	1.00	0.95	0.64	0.88	0.55	0.76	0.80
Janus-Pro	7B	0.99	0.89	0.59	0.90	0.79	0.66	0.80
BAGEL	7B+7B	0.99	0.94	0.81	0.88	0.64	0.63	0.82
HiDream-I1-Full	11B+17B	1.00	0.98	0.79	0.91	0.60	0.72	0.83
UniWorld-V1	7B+12B	0.99	0.93	0.79	0.89	0.49	0.70	0.80
Qwen-Image	7B+20B	0.99	0.92	0.89	0.88	0.76	0.77	0.87
Ovis-Image	2B+7B	1.00	0.97	0.76	0.86	0.67	0.80	0.84

Evaluation of text-to-image generation ability on OneIG-EN.

Model	#Params.	Alignment	Text	Reasoning	Style	Diversity	Overall
Kolors 2.0	-	0.820	0.427	0.262	0.360	0.300	0.434
Imagen4	-	0.857	0.805	0.338	0.377	0.199	0.515
Seedream 3.0	-	0.818	0.865	0.275	0.413	0.277	0.530
GPT4o	-	0.851	0.857	0.345	0.462	0.151	0.533
Ovis-U1	2B+1B	0.816	0.034	0.226	0.443	0.191	0.342
CogView4	6B	0.786	0.641	0.246	0.353	0.205	0.446
Janus-Pro	7B	0.553	0.001	0.139	0.276	0.365	0.267
OmniGen2	3B+4B	0.804	0.680	0.271	0.377	0.242	0.475
BLIP3-o	7B+1B	0.711	0.013	0.223	0.361	0.229	0.307
FLUX.1-dev	11B+12B	0.786	0.523	0.253	0.368	0.238	0.434
BAGEL	7B+7B	0.769	0.244	0.173	0.367	0.251	0.361
BAGEL+CoT	7B+7B	0.793	0.020	0.206	0.390	0.209	0.324
HiDream-I1-Full	11B+17B	0.829	0.707	0.317	0.347	0.186	0.477
HunyuanImage-2.1	7B+17B	0.835	0.816	0.299	0.355	0.127	0.486
Qwen-Image	7B+20B	0.882	0.891	0.306	0.418	0.197	0.539
Ovis-Image	2B+7B	0.858	0.914	0.308	0.386	0.186	0.530

Evaluation of text-to-image generation ability on OneIG-ZN.

Model	#Params.	Alignment	Text	Reasoning	Style	Diversity	Overall
Kolors 2.0	-	0.738	0.502	0.226	0.331	0.333	0.426
Seedream 3.0	-	0.793	0.928	0.281	0.397	0.243	0.528
GPT4o	-	0.812	0.650	0.300	0.449	0.159	0.474
CogView4	6B	0.700	0.193	0.236	0.348	0.214	0.338
Janus-Pro	7B	0.324	0.148	0.104	0.264	0.358	0.240
BLIP3-o	7B+1B	0.608	0.092	0.213	0.369	0.233	0.303
BAGEL	7B+7B	0.672	0.365	0.186	0.357	0.268	0.370
BAGEL+CoT	7B+7B	0.719	0.127	0.219	0.385	0.197	0.329
HiDream-I1-Full	11B+17B	0.620	0.205	0.256	0.304	0.300	0.337
HunyuanImage-2.1	7B+17B	0.775	0.896	0.271	0.348	0.114	0.481
Qwen-Image	7B+20B	0.825	0.963	0.267	0.405	0.279	0.548
Ovis-Image	2B+7B	0.805	0.961	0.273	0.368	0.198	0.521

📚 Citation

If you find Ovis-Image useful for your research or applications, please cite our technical report:

@article{wang2025ovis_image,
  title={Ovis-Image Technical Report}, 
  author={Wang, Guo-Hua and Cao, Liangfu and Cui, Tianyu and Fu, Minghao and Chen, Xiaohao and Zhan, Pengxin and Zhao, Jianshan and Li, Lan and Fu, Bowen and Liu, Jiaqi and Chen, Qing-Guo},
  journal={arXiv preprint arXiv:2511.22982},
  year={2025}
}

🙏 Acknowledgments

The code is built upon Ovis and FLUX. We thank their authors for open-sourcing their great work.

📄 License

This project is licensed under the Apache License, Version 2.0 (SPDX-License-Identifier: Apache-2.0).

🚨 Disclaimer

We used compliance checking algorithms during the training process, to ensure the compliance of the trained model to the best of our ability. Due to complex data and the diversity of language model usage scenarios, we cannot guarantee that the model is completely free of copyright issues or improper content. If you believe anything infringes on your rights or generates improper content, please contact us, and we will promptly address the matter.

🔥 We are hiring!

We are looking for both interns and full-time researchers to join our team, focusing on multimodal understanding, generation, reasoning, AI agents, and unified multimodal models. If you are interested in exploring these exciting areas, please reach out to us at [email protected].