ChatGPT image generation: How it works and how to get the best results

GPT-4o, the latest image generation model from OpenAI, represents a significant advancement in AI technology, replacing the now-outdated DALL·E 3 by incorporating a more sophisticated technique known as visual autoregressive modeling (VAR). Released in 2024 and made available to users in 2025, GPT-4o is a multimodal model capable of handling text, images, audio, and code, making it more powerful due to its extensive training data and deeper understanding of the world. Unlike DALL·E 3's single-pass diffusion method, GPT-4o plans and generates images in stages, allowing for more precise image creation and editing capabilities. Users can generate high-quality images directly through ChatGPT with detailed prompts, and free users can access the feature, albeit with potential limits. GPT-4o enables users to edit uploaded images, change perspectives, and add text, offering a versatile tool for creative projects, although it is not yet a full replacement for traditional photo editors like Photoshop. The model supports natural language follow-ups for further customization and integrates into workflows, although some limitations remain in terms of understanding complex prompts or achieving specific edits.