OpenAI is testing its next-generation image generation models with three variants surfacing on LM Arena and inside ChatGPT A/B tests. Early testers report significant improvements in realistic UI rendering, strong prompt accuracy, and results that are competitive with Google's Nano Banana Pro.
The new models appear to handle complex compositional prompts better than current DALL-E iterations, with particular strength in text rendering within images and maintaining consistent style across multiple generations.
This comes as the image generation space heats up with competition from Midjourney V8, Google's offerings, and a wave of open-source alternatives.