DeepSeek’s Janus Pro B: A New Multimodal AI Model That’s Shaking Things Up

DeepSeek has just released its new model, Janus Pro B, and it’s already making waves in the AI community. This is a multimodal model, meaning it can both generate and understand images—a significant leap forward in AI technology. According to a tweet from Ran Chun, Janus Pro B is not only an upgrade from previous versions but is also outperforming OpenAI’s DALL·E 3 and Stable Diffusion in several benchmarks.

In this article, I’ll walk you through how to use Janus Pro B, compare it against ChatGPT and other AI models, and show you how to get free access to this tool.

The Crazy Timing of DeepSeek’s Success

This major update comes at a wild time—DeepSeek was recently hit by a cyberattack, yet it still managed to overtake ChatGPT on the App Store rankings. According to a tweet from Disclose.tv, DeepSeek AI is now the most downloaded free app in the U.S. App Store, surpassing ChatGPT.

What’s even crazier? The founder of DeepSeek is handling all of this by tweeting memes like an absolute savage. Despite the chaos, the company is pushing forward, and now it’s time to dive into how you can use Janus Pro B yourself.

How to Access and Use Janus Pro B

If you want to try Janus Pro B, you can find it on Hugging Face at:
👉 huggingface.co/spaces/deepseek/janus-pro-7b

Once you’re on the page, you can start generating and analyzing images. This model isn’t just for image generation; it can interpret images as well.

For example, I tested it by asking:
💬 “Explain this meme.”

After tweaking the settings and clicking “Chat,” Janus Pro B generated a response. However, I noticed that it was a bit slow, likely because it’s a brand-new release, and there’s a wait for GPU availability.

⚠️ Running It Locally? I also tried running Janus Pro 7B locally, but it didn’t work at all—it requires a powerful GPU. If you’re thinking of self-hosting from GitHub, I’d say don’t bother unless you have serious hardware.

Janus Pro B vs. ChatGPT: Image Understanding Test

To test its capabilities, I compared Janus Pro B with ChatGPT using the same image.

✅ Janus Pro B’s Response:
It correctly described the meme and its meaning.

✅ ChatGPT’s Response:
ChatGPT also provided a solid explanation.

Verdict: Both AI models performed similarly well in understanding images.

Testing Janus Pro B for Image Generation

Next, I tested Janus Pro B’s image generation by using the prompt:

👉 “AI cat eating biscuits.”

Here’s how it performed compared to other models:

1️⃣ ChatGPT (DALL·E 3) Output

✔️ Image was generated quickly
✔️ Looked a bit weird—half AI, half cat eating some Jammy Dodgers

2️⃣ Janus Pro B Output

✔️ Took longer to generate
❌ Image quality was questionable—biscuits were floating in mid-air, making it look unrealistic

3️⃣ Grok AI Output (My Favorite)

✔️ Generated realistic images
✔️ Fastest generation speed
✔️ No floating biscuits, images actually made sense

🚀 Winner: Grok AI
Grok’s images were by far the best—the most realistic and usable.

Testing More Prompts in Janus Pro B

I also tested a default prompt inside Janus Pro B:

👉 “Master Shifu raccoon wearing drip attire as a street gangster.”

However, I quickly ran into an issue:
⚠️ I exceeded my free GPU quota!

After refreshing the page and trying again, it generated an image, but the quality was not great. The resolution was low, and the details on the face were blurry.

I compared this against ChatGPT and Grok:

ChatGPT refused to generate it (probably due to content policies).
Grok generated a perfect high-quality image.

🚀 Winner: Grok AI again!

The Challenges of Using Janus Pro B

While Janus Pro B is promising, there are some challenges:

Buggy Performance – Hugging Face’s hosting is unstable, and errors pop up frequently.
Slow Generation – Due to high demand, the wait times can be long.
Limited Usability – Image quality isn’t as sharp as Grok’s or DALL·E 3’s.

That said, Janus Pro B is still new, so improvements may come in future updates.

Benchmark Results: How Does Janus Pro B Compare?

Looking at DeepSeek’s own benchmark results, Janus Pro B outperforms many other AI models. However, take these benchmarks with a grain of salt—since they come from DeepSeek itself, they may be biased.

Janus Pro B appears to be stronger than DALL·E 3 and Stable Diffusion in several areas, but real-world results tell a different story.

To test further, I ran this prompt across multiple models:

👉 “Generate a neon billboard with ‘Julian Gold SEO’ written on it.”

Here’s what happened:

Janus Pro B completely failed to generate readable text.
ChatGPT did better, but added extra, unwanted text.
Grok delivered perfect results, making it the clear winner.

🚀 Best AI for Image Generation: Grok AI

Final Verdict: Should You Use Janus Pro B?

🔥 DeepSeek Janus Pro B is a fascinating AI model, and it’s incredible to see how far multimodal AI has come. But at this stage, I wouldn’t fully rely on it for image generation.

Here’s my ranking based on my tests:
✅ 1st Place: Grok AI – Best image generation, fastest speed
✅ 2nd Place: ChatGPT (DALL·E 3) – Good, but limited by content policies
❌ 3rd Place: Janus Pro B – Has potential, but not polished yet

If you want the best image generation AI, Grok AI is still the way to go.