OpenAI launches a new free image creation feature directly in ChatGPT based on the GPT-4o model, which will be available to all users in the near future.
The free usage limit will remain at the DALL-E level, but the company did not specify the specific number of generations, as it will depend on the load. According to the ChatGPT FAQ, previously free users were allowed to generate three images per day using DALL-E 3. However, generation via DALL-E will remain available separately via the corresponding GPT.
Because GPT-4o is an omnimodal model capable of handling text, image, audio, and video, it is better able to recognize complex queries, such as when multiple objects must be represented in a single image. In addition, the quality of text that the system generates directly on images has improved.
The new approach is based on an autoregressive model — the image is created gradually, from left to right and top to bottom, similar to writing text. This distinguishes it from most other models, in particular DALL-E, which use a diffusion technique and generate the entire image at the same time. Although the new system is slower, OpenAI believes that the delay is justified.
The company emphasizes that the system is equipped with effective protection mechanisms against abuse: the removal of watermarks is blocked, the creation of sexual forgeries is prohibited, and CSAM generation requests are rejected.
Images created by the new OpenAI system will not contain visual watermarks or special markers that directly indicate their origin. At the same time, all files will have standard C2PA metadata, which will confirm that they were created with OpenAI.
