DeepSeek Unveils Next-Gen AI Imagery to Compete with Midjourney and DALL-E

DeepSeek's latest AI image generator is a technological breakthrough set to compete with Midjourney and DALL-E. Let's explore what we know so far about this exciting development.

DeepSeek: The New Benchmark in AI

The artificial intelligence sector is rapidly evolving, and the latest buzz is about DeepSeek. Following the success of its open-source DeepSeek R1 model, the company has now unveiled its enhanced image model, Janus Pro, marking a significant event since the launch of ChatGPT in 2022.

Janus Pro: A Formidable Challenger in the AI Arena

Available on GitHub and Hugging Face, Janus Pro comes in two variants: one with one billion parameters and another with seven billion. DeepSeek claims the latter can compete with models like Stable Diffusion and Dall-E 3 in benchmark tests, though independent verification is pending.

DeepSeek’s strength lies in its ability to train models on a significantly smaller budget and older AI chips compared to Western AI models. If these claims hold up and Janus Pro’s performance withstands scrutiny, it could potentially “shake up Silicon Valley”.

Performance and Challenges

Currently, the maximum image size for uploads on Janus Pro is restricted to 384 x 384 pixels. However, according to Android Authority, the Hugging Face demo can generate images up to 768 x 768 pixels. Based on example images from DeepSeek, the Janus Pro 7B represents a significant improvement over the original Janus model.

While we have not yet tested the model ourselves, given the impressive performance of DeepSeek’s chatbot compared to ChatGPT, high expectations surround this new model.

However, DeepSeek has faced its share of issues. Widespread outages have impacted its chatbot’s performance, and alleged “large-scale malicious attacks” have forced the company to temporarily restrict new sign-ups to users with a Chinese phone number.