Alibaba’s Free Picture Era Mannequin is Right here!

Is there one thing Qwen fashions can’t do? To date, their textual content and coding fashions are topping a lot of the charts and arenas. That’s the reason Alibaba’s Qwen workforce obtained onto the “artistic” aspect. They’ve simply launched “Qwen-Picture” – a local textual content rendering picture technology mannequin designed to problem the supremacy of GPT-4.1, DALL-E 2, or Midjourney. The very best half? It’s Free, and what’s even higher is that it’s accessible for everybody! On this weblog, we are going to give you all the main points about Qwen-Picture, together with the way to entry it, its efficiency, functions, and extra. 

Let’s test if the Qwen-Picture is “Qwen-tastic” or not!

What’s Qwen-Picture?

Qwen Picture is the most recent Picture technology mannequin by Alibaba’s Qwen workforce. It’s a 20 B MMDiT picture basis mannequin, that means that the mannequin consists of 20 billion parameters and is a multimodal diffusion transformer mannequin. Qwen-Picture is an open-weight text-to-image technology mannequin that presently ranks fifth on the Synthetic Evaluation Picture Area Leaderboard and is the one open-weight mannequin to be current within the high 10 checklist!

 How does the Qwen-Picture mannequin work?

The Qwen-Picture mannequin follows an method that was final seen in OpenAI’s GPT-4o. It makes use of an autoregressive transformer structure for picture technology and modifying. To do that, the  mannequin takes a twin encoding method: 

• The Qwen2.5-VL encodes the semantic that means of the immediate
• Picture technology occurs in a latent house utilizing MMDiT, a diffusion mannequin
• The ultimate picture is produced from this latent house utilizing a VAE encoder. 

You possibly can learn the complete technical report of the Qwen-Picture mannequin right here.

Key Options of Qwen-Picture

A few of the key highlights that make Qwen-Picture stand aside are:

1. Enhanced Textual content Incorporation: The Qwen-Picture fashions are distinctive relating to incorporating complicated texts, whether or not in multi-line layouts, paragraphs, and even fine-grained particulars. It really works equally nicely with each alphabetic languages (reminiscent of English) and logographic languages (like Chinese language), with the identical ease. 
2. Environment friendly Picture Modifying: The mannequin affords superior picture modifying capabilities. In the course of the modifying course of, the mannequin preserves each the semantic and visible that means of the particular photographs whereas incorporating the brand new adjustments. 
3. Ease of Use: The mannequin is straightforward to make use of and works nicely even with easy prompts. 

These options, together with the superb efficiency of this mannequin, have been showcased on numerous benchmarks- making Qwen-Picture a formidable picture technology mannequin.

The way to entry Qwen-Picture?

To entry the Qwen-Picture mannequin by means of Chat, 

1. Head to https://chat.qwen.ai/
2. Choose any of the non-coding fashions like Qwen-235B-A3B-2507 

3. Beneath the textual content field, in the midst of the display screen, choose “Picture Era”

Enter your immediate within the textual content field and get began!

You possibly can entry the fashions in different methods, like:

Qwen-Picture: Handson

Now that we have now lined a number of particulars about Qwen-Picture, let’s take a look at it for 3 foremost duties:

1. Producing a text-heavy Picture
2. Producing an Infographic
3. Modifying an Picture

Let’s begin with every of them one after the other:

Activity: 1: Design a Net Web page

Immediate: “Create a visually partaking touchdown web page for a shampoo product. Spotlight the shampoo’s distinctive options (e.g., hydration, restore, or pure substances) with a clear and fashionable design. Embrace a hero part with the shampoo bottle picture, a catchy headline like ‘Remodel Your Hair At present,’ and a call-to-action button (‘Store Now’ or ‘Be taught Extra’). Add sections for advantages, key substances, buyer testimonials, and a subscription possibility. Use mushy, recent colours, high-quality visuals, and make sure the format is mobile-friendly and conversion-focused.”

Output:



The generated picture was good; it had a number of the textual content that I had requested to be integrated. It captured the essence of the immediate nicely and designed the complete picture appropriately. However there have been just a few misses. Though spellings had been right, at one place a phrase was incomplete, and a few phrases that I had talked about weren’t integrated. I preferred the color theme that the mannequin selected for this activity.

Activity 2: Create a Flowchart

Immediate: “ Design a transparent, fashionable infographic that explains the picture technology strategy of a 20B MMDiT basis mannequin in 3 steps:

• Immediate Encoding: Present Qwen2.5-VL encoding the semantic that means of the consumer’s immediate.
• Latent House Era: Visualize MMDiT diffusion creating an summary picture in latent house.
• Ultimate Picture Creation: Illustrate a VAE decoder remodeling the latent illustration into the ultimate high-quality picture.

Use icons, arrows, and quick labels for every step. The movement ought to be visually logical and simple to observe, with a tech-inspired shade palette.”

Output:



I didn’t just like the output in any respect. The textual content was lacking in some locations and utterly obscure at different locations. The icons and total picture felt a bit disoriented. The movement from step 1 to 2 to three was there, however the picture is sort of unclear. 

Activity 3: Picture Modifying

Enter picture:



Immediate: “Change the evening right into a sunny morning, exchange the person’s garments with an orange shirt and white shorts, and exchange the cat with a small pet.”

Output:



This consequence was simply good. Actually Excellent. All of the adjustments that I had requested for occurred within the picture. The lighting was appropriate, the garments and the animal had been all modified. A minor challenge: whereas the mannequin changed evening with day, it didn’t take away the moon, though it made it seem like a spherical cloud. A really nicely edited picture that took only a few seconds to generate!

My Evaluate Utilizing Qwen-Picture

General, I actually preferred the modifying capabilities of the mannequin, however the picture technology, particularly incorporating a considerable amount of textual content or designing infographics, is the place Qwen-Picture would want a number of enchancment going ahead – particularly if it needs to compete with the likes of OpenAI, Google, or X. 



However it has one actually cool characteristic that a lot of the high fashions don’t. You possibly can really choose the body dimension that you just want to work with, proper from the textual content field! In case you are a content material creator, this actually would enable you to create the “right-sized” picture for every of your social media platforms.

Qwen Picture: Efficiency 

Now that we have now examined the mannequin, let’s have a look at the outcomes that the Qwen workforce has launched for the efficiency of the Qwen-Picture mannequin in opposition to its counterparts:

1. For Picture Era and Modifying Benchmarks



• Qwen-Picture mannequin leads or is at par with the perfect fashions in nearly all of the picture technology & modifying benchmarks. 
• GPT-4.1 and Seedream3.0 are shut rivals of Qwen-Picture, matching its scores on a number of benchmarks.
• FLUX.1 fashions are a great competitors however lag behind the Qwen-image mannequin 

2. For Textual content Rendering Benchmarks:



• Qwen-Picture leads for textual content rendering in Chinese language and can also be fairly forward for English languages
• GPT4.1 – surpasses or matches Qwen-image at numerous benchmarks. 
• Seeddream 3.0 is a detailed competitor however lags behind Qwen-Picture in each Chinese language and English benchmarks. 

Conclusion:

Qwen fashions are presently ruling the leaderboards for textual content and coding-based duties. Qwen-Picture holds related promise however isn’t fairly there but. The mannequin adheres to prompts however struggles with big context. However it’s a fantastic present to the open-source group. It competes with the top-paid fashions whereas being utterly open-weight. As customers and builders use Qwen-Picture increasingly, we are able to quickly count on the Qwen-Picture mannequin to steer the Picture Era Evaluation too!

My closing thought – attempt the Qwen-Picture Mannequin. It’s good, we’re simply surrounded by a number of nice fashions to not realise its potential. 

You may as well examine Discovering the Finest AI Picture Era Mannequin.

If you wish to examine different FREE picture technology fashions, you may confer with the next weblog: Prime 7 AI Picture Mills to Attempt in 2025.




Anu Madan is an professional in tutorial design, content material writing, and B2B advertising and marketing, with a expertise for remodeling complicated concepts into impactful narratives. Together with her deal with Generative AI, she crafts insightful, revolutionary content material that educates, evokes, and drives significant engagement.