Beware of ChatGPT Images 2.0 Deceptions!

On April 22, OpenAI officially launched the highly anticipated image generation and editing model, ChatGPT Images 2.0, which is now fully integrated into the ChatGPT chatbot and API services.

Sam Altman described the release as a significant leap, stating, “The launch of ChatGPT Images 2.0 is like jumping directly from GPT-3 to GPT-5.”

After its launch, the model quickly topped all rankings on Image Arena, outperforming competitors like Nano banana 2 by a staggering 242 points.

The release sparked a frenzy on social media, with users exclaiming, “Bananas are no longer appealing; ChatGPT Images 2.0 has killed the competition.”

Many users showcased images generated by ChatGPT Images 2.0, leading to a flood of stunning visuals that blur the line between reality and fabrication. Some even discovered that it could generate scannable barcodes.

Among the showcased results were live stream screenshots, conference photos, and even transfer records, all appearing incredibly realistic and closely tied to everyday life. These images generated by ChatGPT Images 2.0 can easily deceive unsuspecting individuals.

Live Streams, Transfers, and Conferences: AI-Generated Images Are Hard to Distinguish

The announcement image of DeepSeek V4 was generated using ChatGPT Images 2.0, with the prompt being: “A screenshot of a tweet announcing the release of DeepSeek V4, published by the official DeepSeek account, retaining elements like avatar, username, likes, retweets, and comments.”

Additionally, ChatGPT Images 2.0 generated a screenshot of a WeChat Moments post, making the situation appear more cohesive. The prompt was: “A WeChat Moments screenshot where Sam Altman praises DeepSeek V4 as a groundbreaking model, with likes and comments from Musk, Zuckerberg, and Huang Renxun below.”

Without reference images, the screenshots generated by ChatGPT Images 2.0 appeared quite realistic, although the generated avatars had some discrepancies, and Huang Renxun’s name was slightly misspelled.

The model also generated a live stream page, a trend that has recently gone viral during internal testing. Someone created a screenshot of Altman promoting a product during a live stream, which looked entirely believable, complete with a prize draw and realistic comments. If this were replaced with the image of another well-known figure, many might be deceived.

ChatGPT Images 2.0 was also tasked with generating a photo of handwritten notes. The prompt was: “Take a photo of a high school math assignment written by hand, with neat handwriting on white paper, featuring a complex derivative problem.”

The generated photo resembled many neatly written student assignments, prompting comments to see if teachers could distinguish the fake.

Some even created scenarios where Steve Jobs announced a Xiaomi car, and upon close inspection, it was hard to identify any discrepancies.

If the previous image didn’t seem realistic enough, consider the following:

This image, also generated by ChatGPT Images 2.0, had the prompt: “At the Apple Park, Tim Cook is giving a speech at the iPhone 20 launch event. This amateur iPhone photo was taken from a distance in the crowd.”

The photo matched a typical media perspective, and if it weren’t for the glaring iPhone 20 and Cook’s retirement, it could easily mislead many.

Next, we see an academic paper generated in ChatGPT, with the prompt: “A page from the Peking University Journal featuring a paper on AI safety issues, beautifully typeset in a rigorous style.”

The academic journal page was well-crafted, adhering closely to the standards of formal academic publications. The layout was clear, with accurate Chinese text, and even included DOI and copyright information, making it look very legitimate. If someone shared this paper as a reference, many might be fooled.

Even transfer screenshots were generated using ChatGPT Images 2.0, raising concerns as they could even forge official seals, which is alarming.

This prescription image was also created by ChatGPT Images 2.0, and aside from the overly neat handwriting that doesn’t resemble a doctor’s, everything else could easily deceive.

GPT Image 2.0 Receives Rave Reviews, Users Are Going Wild

Currently, all ChatGPT and Codex users can use a limited number of image generations for free, with tests showing around seven images can be generated. Paid subscribers can unlock a thinking mode, allowing the model to perform online searches, tests, and self-checks to ensure image quality, and the gpt-image-2 model has been opened to external APIs.

Users can open a ChatGPT conversation, click the plus sign next to the input box to create images, which will default to using ChatGPT Images 2.0. They can then add photos and files and send prompts.

After the release of GPT Image 2.0, users eagerly tried it out, praising its capabilities, with many showcasing the stunning images they generated. Some users exclaimed that it was nearly impossible to distinguish between real and fake.

A medical doctor from the University of Tokyo shared a poster of a paper he generated, stating that although he hadn’t tested it yet, ChatGPT Images 2.0 seemed capable of handling data in academic papers, exclaiming that scientific research is about to undergo another transformation.

Riley Brown, co-founder of vibecode and an overseas blogger, stated, “This is the best image model.”

While ChatGPT Images 2.0 is incredibly powerful, many deceptive images still require vigilance. The ethical issues surrounding AI-generated images cannot be overlooked. If these convincingly fake images were shared by familiar people or companies on social media, how many would fall for them?

Conclusion: AI Image Generation Reaches New Heights, but Requires Regulation

With the increasing demand for AI-generated images, the release of ChatGPT Images 2.0 undoubtedly elevates the entire industry to a new level. The clarity of generated images, detail restoration capabilities, style diversity, and creative expression freedom have all seen significant improvements.

However, with enhanced technological capabilities, the industry faces new challenges, including copyright protection, content review, and ethical risks associated with generated content. Many of these AI-generated “masterpieces” can easily deceive, such as conference announcements, WeChat Moments screenshots, and live stream images. If such content is flooded with AI-fabricated images, how many people will be deceived?

Future developments will depend not only on algorithmic advancements but also on standardized industry norms and sustainable application strategies.

As technology and governance improve in tandem, AI-generated images will play a more significant role in creative design, education, and entertainment.