DeepSeek’s Janus Pro AI Image Generator Takes on DALL-E and Midjourney

DeepSeek, the viral Chinese AI company, has launched Janus Pro, a new family of multimodal AI models designed to analyze and generate images. The company claims that Janus Pro 7B, its largest model, outperforms OpenAI’s DALL-E 3 as well as PixArt-alpha, Emu3-Gen, and Stable Diffusion XL on key AI evaluation benchmarks.

Unlike many commercial AI image generators, Janus Pro is open-source under an MIT license, meaning it can be used without restrictions for both personal and commercial projects.

What Sets Janus Pro Apart?

Janus Pro models range from 1 billion to 7 billion parameters—a relatively compact size compared to many competing models. Despite this, DeepSeek claims they deliver high-quality image generation and analysis using a novel autoregressive framework, which can also create new images.

Key features of Janus Pro:

• Multimodal Capabilities: Can both analyze existing images and generate new ones.

• Open-Source & Commercially Usable: Available on Hugging Face under an MIT license.

• Competitive Performance: Beats DALL-E 3 and other models on GenEval and DPG-Bench benchmarks.

• Compact Yet Powerful: Achieves high performance with fewer computational resources than many competitors.

• Resolution Limits: Currently limited to 384 x 384 pixels, making it less suitable for high-resolution artwork.

You can visit HuggingFace to see the details.

Implications for AI Image Generation

DeepSeek’s rise has sparked industry discussions about the future of AI development, particularly as its models use compute-efficient techniques to achieve strong results with fewer resources. With the demand for AI chips skyrocketing, models like Janus Pro could influence how AI companies balance efficiency and performance in the coming years.

Despite its early limitations, Janus Pro’s open-source nature makes it a viable competitor to closed systems like DALL-E, Midjourney, and Adobe Firefly. If DeepSeek continues improving the model’s resolution and scalability, it could further disrupt the AI-generated image space.

DeepSeek’s rapid development of competitive AI models—first with R1 and now with Janus Pro—has also raised questions about how the company is achieving such breakthroughs so quickly. Given Microsoft’s ongoing investigation into whether DeepSeek improperly extracted OpenAI API data to train R1, some may wonder whether similar concerns apply to Janus Pro. However, no direct evidence has emerged linking Janus Pro to unauthorized training data.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.