With the recent release of the DeepSeek AI model, everyone is looking for its image generation tool. Some people created fake websites named DeepSeek Image Generation and charged high prices for generating images. However, In this article, I will clarify whether really DeepSeek generates images or not. I will proceed to the point answer to this question.
Can DeepSeek Generate Images
Yes, DeepSeek’s Janus-Pro can generate the images, even claims that beating OpenAI’s DALL-E 3 and Stability AI’s Stable Diffusion in benchmark tests, placing it among its competitors. However, they didn’t have a dedicated website or app to generate the images, have been launched on Free sources like GitHub and Hugging Face, available in both one-billion and seven-billion parameter models. It hasn’t been thoroughly tested by third parties, can’t clarify the accuracy and quality of the image.
Image Generation Size
The image size is restricted to 384 x 384 pixels as of now, but the Hugging Face demo can generate 768 x 768 pixel images. DeepSeek shares example images of the Janus-Pro-7B significant upgrade compared to its old model Janus. Check out the below images.
Can It Read Images
Yes, it can read the images and provide data, and image information. Also analyzes the objects, and colors and interprets them in texts.
Can It edit Images
No, it can’t edit the images. For Example. Microsoft Copilot can edit and remove the background in the images as you want.
Can It Search Images
Yes, It can search images as an in-built “Search” button for search for images or anything.
How To Use DeepSeek Janus Pro
To use the DeepSeek Janus Pro image generator follow the below steps:
Step 1:
First Visit Hugging Face and click on the search button, search deepseek-ai/Janus-Pro-7B.
Step 2:
Scroll below and locate the space using the section, you will this list
Step 3:
Click on the LLMhacker/DeepseekJanusPro-Image and it will be redirected to another page.
Step: 4
you get this interface and enter the prompt image you want to generate.
Results
Prompt “Design a futuristic cyberpunk cityscape at night, with towering neon skyscrapers, flying cars, and bustling streets filled with diverse characters. The scene should be highly detailed, with a focus on reflective surfaces, glowing neon signs, and a mix of vibrant colors like electric blue, hot pink, and deep purple. Use a hyper-realistic art style with strong contrasts between light and shadow. The mood should be energetic yet gritty, capturing the essence of a high-tech, dystopian world.”
This is the result of the HuggingFace Demo Janus-Pro, it seems like garbage and I will put the model with Midjourney V2, they compared the model with DALL-E 3 and Stable Diffusion outdated versions instead of newer and better image models, which’s not a fair comparison.
Final Thoughts
While DeepSeek Janus-Pro shows some promise in AI-generated images, it’s far from being a real competitor to DALL-E 3 or even the latest versions of Midjourney. The model’s small image size, lack of proper third-party testing, and questionable benchmark comparisons make it clear that it’s still in an early stage. If you’re looking for serious AI-generated art, this isn’t the best option yet. However, since it’s open-source and evolving, it might improve in the future. But for now, the hype doesn’t match the reality.
I’m Ryker Alden, a writer, and contributor at DeepSeek Insider, where I craft queries, troubleshoot problems and create accessible tutorials. With a passion for artificial intelligence, machine learning, and large language models (LLMs), I focus on breaking down complex AI concepts into clear, simple language to engage and educate a broad audience.