graphic designers are cooked worse than programmers.     See Less

1:00 The shape of their hands appear to be copy pasted     See Less

Summary (with chapter timestamps): OpenAI's new ImageGen excels in obedience, complex prompts, and logical understanding, outperforming competitors like Reeve and Imagen 3, making it a t     See More in AI image generation.
0:00 ✨ Introduction to OpenAI's New ImageGen
• OpenAI's new image generation model (40 image gen) is surprisingly effective and obedient compared to other models.
• It will be available to everyone, even free users, in chat GPT and the API.

1:04 🐘 Testing with a Complex Prompt
• The model accurately rendered a complex prompt: three apples balanced on a blue elephant with three legs, standing beside five weeping willow trees in El Jem, Tunisia.
• While it struggled with the three legs, it captured the location, apples, and trees effectively, even maintaining consistent shadows.
• Imagen 3 struggled with the apples and location, while Reeve produced a good image but occasionally got the number of apples wrong.

3:35 🐴 Understanding Idioms
• 40 image gen understood the idiom 'hold your horses' and conveyed it appropriately in every image.
• Reeve and Imagen 3 couldn't grasp the metaphor, showcasing 40 image gen's superior understanding of abstract concepts.

4:18 🖼️ Thumbnail Generation and Image Editing
• 40 image gen can transform classic thumbnails into 3D versions with impressive accuracy, including text.
• It can also generate lab-like thumbnails for videos, tempting the creator to use AI-generated thumbnails.
• The model effectively creates images with captions or basic infographics, such as depicting the stages of human life with labels.

5:59 👓 Image Editing Capabilities
• Image editing is seamless with chat GPT, allowing users to easily add glasses to characters in an image.
• Other image generators struggled with depicting the four stages of life, while 40 image gen performed well.
• Google AI studio with Gemini 2 flash produced less favorable results when depicting the stages of human life.

7:20 🛡️ Filtering and Safety
• The model allowed the submission of a photo of the Google CEO and Sam Altman arm wrestling, suggesting slightly less filtering.
• The creator encourages viewers to participate in the grace one Arena to test model vulnerabilities, including visual ones.
• 40 image gen excels at producing images with accurate text and genuine logic, marking a significant moment in AI.

9:11 🔍 Logic and Artisticness
• 40 image gen's logic and reasoning are noticeably better than other models, while its artisticness is comparable.
• The model can create 'Where's Wally' style images, although it sometimes struggles to include the item being searched for.
• It can also turn selfies into charcoal sketches or Dragon Ball Z characters.


** Generated using ✨ VidSkipper AI Chrome Plugin    See Less

i think for image gen, you should have compared models with native generation such as 4o vs flash     See Less

You should have added labels so we could see which image came from what model.     See Less

Testing the most obscure edge cases is not very useful imo     See Less

Didn't you see the pirate? 😂     See Less

The elephant does have three legs. It also has a fourth.     See Less

its dalle3. you can do that with bing     See Less

As a prominent AI image generating user, hands haven't been really a problem for over 8 months sir! But yes these examples are good!     See Less