Summary
- OpenAI’s GPT-4o model allows users to generate images and restyle photos, but may soon add watermarks for free-tier users.
- Watermarking AI-generated media is a hot topic, yet OpenAI is testing watermarks only for images created by free-tier users.
- ChatGPT’s GPT-4o release offers image generation, but the watermark may be used to encourage users to subscribe to the premium plans.
OpenAI focused its efforts on popularizing the chatbot experience ChatGPT offered since the initial public launch. At the time, image generation was handled by specialized AI tools like Midjourney. A year later, the boundaries are blurred because chatbots can create simple images too. Meta AI and Google’s Gemini do it, but OpenAI is rather new on the scene with its GPT 4o model. Now, the company is reportedly testing watermarking for generated images.
Related
11 AI tools I can’t believe are free
These AI tools cost nothing, which is exactly how much I want to spend
GPT-4o was released to the public for free last week, after a short period where only paid subscribers could access it. Besides several other improvements, this new model unlocks image generation for users, you can supply a fully textual prompt or use an existing image and have it restyled. The internet is abuzz with ChatGPT restyling photos to match Studio Ghibli’s art style as well. However, to limit the toll of a surge in usage, free users are likely limited to generating only two images per day.
However, rampant usage of the Studio Ghibli art style seems to have highlighted another problem with GPT-4o — lack of watermarking. AI researcher Tibor Blaho, who goes by @btibor91 on X, recently shared that version 1.2025.0912509108 of the ChatGPT Android app mentions a new ImageGen watermark (via BleepingComputer). ImageGen is the model that powers the image generation capabilities of the multi-model experience that GPT-4o is.
Watermarks only on free-tier output?
An odd choice
While the AI industry is collectively pushing for watermarking AI-generated media so it is never usable as a substitute for human-created content, OpenAI is doing something weird by testing watermarking only for images generated by free-tier users. This could resemble how stock image distributors watermark free download copies and demand payment for larger resolution or watermark-free results, serving as a lure for ChatGPT’s paid tiers.
This is likely separate from the digital watermarking that distinguishes human work from the AI’s output, but it will be interesting to see how users take the change when it rolls out. OpenAI hasn’t mentioned this publicly or committed to a timeline yet, but rivals like Gemini and Meta AI already add visible watermarks to generated images.
