DeepSeek has made waves in the AI market by introducing a low-cost ChatGPT competitor with similar capabilities. Now, the company is venturing into image generation with its new open-source model, Janus Pro. According to tests shared by the startup, the results are promising and competitive with existing industry leaders.
The Janus Pro 7B model, launched on Monday (27), features multimodal functionality, allowing it to interpret multiple media formats. DeepSeek also highlighted improvements in image generation capabilities. Benchmark tests released by the company indicate that Janus Pro 7B outperforms OpenAI’s DALL-E 3 and older versions of Stable Diffusion models in key performance metrics.
It’s important to note that these benchmark tests don’t necessarily evaluate the final quality of the generated images. Instead, they focus on the model’s overall performance and its ability to understand and execute user commands effectively.
How Janus Pro works
The Janus Pro 7B model is currently in its demonstration phase but can already be tested on the Hugging Face platform. The functionality is straightforward: users input a text prompt describing their desired image and receive four generated samples.
That said, the tool on Hugging Face does have some limitations. Images are relatively small, and there are noticeable delays in generating responses. However, these issues are likely tied to the demonstration setup and not the model’s final version.
In terms of results, Janus Pro still lags behind leading market competitors and feels more akin to the early versions of similar AI models. For instance, in a test by Canaltech using the prompt “a dog on the streets of Rio de Janeiro,” Janus Pro generated a generic background with only a silhouette of the dog. In contrast, DALL-E produced a far more detailed and colorful image, showing its edge in image quality and refinement.
All Janus Pro files and documentation are available in the DeepSek Github Repository (github.com/deepseek-ai/janus).
What is Deepseek?
DeepSeek is an AI chatbot developed by a Chinese startup of the same name. Launched in January this year, it features the DeepSeek R1 model, which quickly gained attention for delivering results comparable to ChatGPT but with significantly lower infrastructure costs.
The debut of this Chinese AI shook the tech industry, particularly impacting the stock market of major players. Nvidia, a leader in advanced AI chips, experienced a dramatic 17% drop in market value in a single day, with an estimated loss of $580 billion in share value. This underscores the disruptive potential of DeepSeek in the AI landscape.
