Published in Tech

Image credit by Argo

Sophie

March 27, 2025

🔥 OpenAI's GPT-4o revolutionises image generation: More beautiful, more accurate, and downright stunning!

OpenAI has launched "4o Image Generation", an image generator integrated with GPT-4o that creates functional visuals, not just aesthetic ones. The model excels at rendering precise text, following complex instructions, maintaining multi-turn coherence, contextual learning, and utilizing embedded knowledge. Capable of impressive photorealism and varied styles, this technology is already available to ChatGPT users, despite some limitations such as image cropping or difficulties with non-Latin languages.

OpenAI has just dropped a real bombshell in the world of image generation, and honestly, it's huge. On March 25, 2025, they unveiled "4o Image Generation", their new image generator integrated directly into GPT-4o. And be warned, we are not talking about a simple update - this is a full-blown revolution!

📱 No more just "pretty" images - it's time for truly USEFUL images

We've all seen AI generating beautiful sunsets or fantasy portraits, but struggling when it comes to doing something precise, right? Well, that's in the past! OpenAI has understood that from cave paintings to modern infographics, images have not just served to decorate but to communicate, persuade, and analyze.

As they say so well: "A picture is worth a thousand words, but sometimes a few words placed in the right spot can elevate the meaning of an image." And that perfectly sums up their approach!

💪 The superpowers of the new model

1. Finally perfect text management

No more weird or unreadable text in your generated images! This model excels at rendering precise text. Want a wedding invitation with perfectly legible text? An educational infographic with clear captions? A stylish restaurant menu with all the correct descriptions? It's now possible!

Prompt used: "Create an elegant marketing image for ARGO showing a fashion advertisement in a magazine that 'comes to life'. The image should show a model in the printed page that seems to be emerging from the page through augmented reality when viewed through a smartphone. Make sure the image on the smartphone is perfectly aligned with the printed image, but animated. Include the ARGO logo and a small text saying 'Increase the impact of your customer communications'. Photorealistic style, bright and professional."

2. Accurate instruction following

You can now give incredibly detailed instructions and the model will follow them to the letter. While other models struggle with 5-8 objects, GPT-4o can handle 10-20 different objects along with their specific relationships and attributes!

3. Multi-turn generation that maintains consistency

The model remembers previous images and maintains consistency. You can refine your image through a natural conversation without losing important details. Imagine creating a video game character and being able to gradually adjust its appearance while keeping its distinctive features!

4. Contextual learning

It can analyze uploaded images and draw inspiration from them for new creations. Show it a sketch, and it can transform it into a realistic image or adapt it to another style!

5. Integrated knowledge

The model uses all its knowledge to create informative and precise images. Ask it for an infographic on fog in San Francisco or an educational poster about whales, and it will know exactly what to include!

🤩 Examples that rock

  • A mini comic with character consistency and logo integration

  • Image illustrating the benefits of ARGO technology

  • Graph for explaining augmented reality




🎨 Breathtaking photorealism

The model also excels in photorealism and various artistic styles. From the comical paparazzi portrait of Karl Marx at the mall to surreal underwater scenes with dolphins swimming through the windows of an abandoned subway car, the possibilities are endless!

🔒 Enhanced security

OpenAI has not neglected security. All generated images are tagged with C2PA metadata to ensure transparency. The system blocks inappropriate requests and uses a "reasoning LLM" to enforce security policies, similar to their approach to "deliberative alignment".

🤷‍♂️ A few limitations (at least they are honest)

The model is not perfect. It can sometimes:

  • Crop long images too tightly like posters

  • Invent information (hallucinate) in prompts with little context

  • Struggle with more than 10-20 distinct concepts at once

  • Have difficulty with text in non-Latin languages

  • Lack precision when editing specific portions of an image

🚀 Where to try it out?

The good news is that 4o image generation is now rolled out for Plus, Pro, Team users, and even free ChatGPT users as the default image generator! Enterprise and Edu users will have access soon. It is also available in Sora.

Developers will be able to generate images with GPT-4o via the API in the coming weeks. And for those nostalgic for DALL·E, don’t worry - it remains accessible via a GPT dedicated to DALL·E.

👀 The final word

OpenAI's new image generator is not just a toy for making pretty images - it's a real visual communication tool. It brings image generation closer to what humans have been doing for millennia: using images to share ideas, convey information, and tell stories.

So, ready to try it out? Images can take up to a minute to generate (hey, quality has a price!), but the result is definitely worth it!

This article was generated using information from official OpenAI publications on March 25, 2025.

OpenAI has just dropped a real bombshell in the world of image generation, and honestly, it's huge. On March 25, 2025, they unveiled "4o Image Generation", their new image generator integrated directly into GPT-4o. And be warned, we are not talking about a simple update - this is a full-blown revolution!

📱 No more just "pretty" images - it's time for truly USEFUL images

We've all seen AI generating beautiful sunsets or fantasy portraits, but struggling when it comes to doing something precise, right? Well, that's in the past! OpenAI has understood that from cave paintings to modern infographics, images have not just served to decorate but to communicate, persuade, and analyze.

As they say so well: "A picture is worth a thousand words, but sometimes a few words placed in the right spot can elevate the meaning of an image." And that perfectly sums up their approach!

💪 The superpowers of the new model

1. Finally perfect text management

No more weird or unreadable text in your generated images! This model excels at rendering precise text. Want a wedding invitation with perfectly legible text? An educational infographic with clear captions? A stylish restaurant menu with all the correct descriptions? It's now possible!

Prompt used: "Create an elegant marketing image for ARGO showing a fashion advertisement in a magazine that 'comes to life'. The image should show a model in the printed page that seems to be emerging from the page through augmented reality when viewed through a smartphone. Make sure the image on the smartphone is perfectly aligned with the printed image, but animated. Include the ARGO logo and a small text saying 'Increase the impact of your customer communications'. Photorealistic style, bright and professional."

2. Accurate instruction following

You can now give incredibly detailed instructions and the model will follow them to the letter. While other models struggle with 5-8 objects, GPT-4o can handle 10-20 different objects along with their specific relationships and attributes!

3. Multi-turn generation that maintains consistency

The model remembers previous images and maintains consistency. You can refine your image through a natural conversation without losing important details. Imagine creating a video game character and being able to gradually adjust its appearance while keeping its distinctive features!

4. Contextual learning

It can analyze uploaded images and draw inspiration from them for new creations. Show it a sketch, and it can transform it into a realistic image or adapt it to another style!

5. Integrated knowledge

The model uses all its knowledge to create informative and precise images. Ask it for an infographic on fog in San Francisco or an educational poster about whales, and it will know exactly what to include!

🤩 Examples that rock

  • A mini comic with character consistency and logo integration

  • Image illustrating the benefits of ARGO technology

  • Graph for explaining augmented reality




🎨 Breathtaking photorealism

The model also excels in photorealism and various artistic styles. From the comical paparazzi portrait of Karl Marx at the mall to surreal underwater scenes with dolphins swimming through the windows of an abandoned subway car, the possibilities are endless!

🔒 Enhanced security

OpenAI has not neglected security. All generated images are tagged with C2PA metadata to ensure transparency. The system blocks inappropriate requests and uses a "reasoning LLM" to enforce security policies, similar to their approach to "deliberative alignment".

🤷‍♂️ A few limitations (at least they are honest)

The model is not perfect. It can sometimes:

  • Crop long images too tightly like posters

  • Invent information (hallucinate) in prompts with little context

  • Struggle with more than 10-20 distinct concepts at once

  • Have difficulty with text in non-Latin languages

  • Lack precision when editing specific portions of an image

🚀 Where to try it out?

The good news is that 4o image generation is now rolled out for Plus, Pro, Team users, and even free ChatGPT users as the default image generator! Enterprise and Edu users will have access soon. It is also available in Sora.

Developers will be able to generate images with GPT-4o via the API in the coming weeks. And for those nostalgic for DALL·E, don’t worry - it remains accessible via a GPT dedicated to DALL·E.

👀 The final word

OpenAI's new image generator is not just a toy for making pretty images - it's a real visual communication tool. It brings image generation closer to what humans have been doing for millennia: using images to share ideas, convey information, and tell stories.

So, ready to try it out? Images can take up to a minute to generate (hey, quality has a price!), but the result is definitely worth it!

This article was generated using information from official OpenAI publications on March 25, 2025.

OpenAI has just dropped a real bombshell in the world of image generation, and honestly, it's huge. On March 25, 2025, they unveiled "4o Image Generation", their new image generator integrated directly into GPT-4o. And be warned, we are not talking about a simple update - this is a full-blown revolution!

📱 No more just "pretty" images - it's time for truly USEFUL images

We've all seen AI generating beautiful sunsets or fantasy portraits, but struggling when it comes to doing something precise, right? Well, that's in the past! OpenAI has understood that from cave paintings to modern infographics, images have not just served to decorate but to communicate, persuade, and analyze.

As they say so well: "A picture is worth a thousand words, but sometimes a few words placed in the right spot can elevate the meaning of an image." And that perfectly sums up their approach!

💪 The superpowers of the new model

1. Finally perfect text management

No more weird or unreadable text in your generated images! This model excels at rendering precise text. Want a wedding invitation with perfectly legible text? An educational infographic with clear captions? A stylish restaurant menu with all the correct descriptions? It's now possible!

Prompt used: "Create an elegant marketing image for ARGO showing a fashion advertisement in a magazine that 'comes to life'. The image should show a model in the printed page that seems to be emerging from the page through augmented reality when viewed through a smartphone. Make sure the image on the smartphone is perfectly aligned with the printed image, but animated. Include the ARGO logo and a small text saying 'Increase the impact of your customer communications'. Photorealistic style, bright and professional."

2. Accurate instruction following

You can now give incredibly detailed instructions and the model will follow them to the letter. While other models struggle with 5-8 objects, GPT-4o can handle 10-20 different objects along with their specific relationships and attributes!

3. Multi-turn generation that maintains consistency

The model remembers previous images and maintains consistency. You can refine your image through a natural conversation without losing important details. Imagine creating a video game character and being able to gradually adjust its appearance while keeping its distinctive features!

4. Contextual learning

It can analyze uploaded images and draw inspiration from them for new creations. Show it a sketch, and it can transform it into a realistic image or adapt it to another style!

5. Integrated knowledge

The model uses all its knowledge to create informative and precise images. Ask it for an infographic on fog in San Francisco or an educational poster about whales, and it will know exactly what to include!

🤩 Examples that rock

  • A mini comic with character consistency and logo integration

  • Image illustrating the benefits of ARGO technology

  • Graph for explaining augmented reality




🎨 Breathtaking photorealism

The model also excels in photorealism and various artistic styles. From the comical paparazzi portrait of Karl Marx at the mall to surreal underwater scenes with dolphins swimming through the windows of an abandoned subway car, the possibilities are endless!

🔒 Enhanced security

OpenAI has not neglected security. All generated images are tagged with C2PA metadata to ensure transparency. The system blocks inappropriate requests and uses a "reasoning LLM" to enforce security policies, similar to their approach to "deliberative alignment".

🤷‍♂️ A few limitations (at least they are honest)

The model is not perfect. It can sometimes:

  • Crop long images too tightly like posters

  • Invent information (hallucinate) in prompts with little context

  • Struggle with more than 10-20 distinct concepts at once

  • Have difficulty with text in non-Latin languages

  • Lack precision when editing specific portions of an image

🚀 Where to try it out?

The good news is that 4o image generation is now rolled out for Plus, Pro, Team users, and even free ChatGPT users as the default image generator! Enterprise and Edu users will have access soon. It is also available in Sora.

Developers will be able to generate images with GPT-4o via the API in the coming weeks. And for those nostalgic for DALL·E, don’t worry - it remains accessible via a GPT dedicated to DALL·E.

👀 The final word

OpenAI's new image generator is not just a toy for making pretty images - it's a real visual communication tool. It brings image generation closer to what humans have been doing for millennia: using images to share ideas, convey information, and tell stories.

So, ready to try it out? Images can take up to a minute to generate (hey, quality has a price!), but the result is definitely worth it!

This article was generated using information from official OpenAI publications on March 25, 2025.