While talking about AI, in an alternate dimesion where the world is fair even taking a single word from a book for their AI models without asking permissions to their authors would be considered a crime, the same thing with art, even in that reality they would eventually gotten access by slowly and steadily buying new books and art pieces but that would mean a gradual change, in such a slow manner that society would have time to adapt, not like this reality we have now.

I know how it goes, “just one more thing you need to learn to accept” is a given in this chaotic universe and reality, its easy to take that in when we are talking about things like cancer or viruses or other random diseases, but when they are arbitrary decisions by a handful of randomly selected people is different, like there was a shot at making this tiny part of existence a positive experience and with every single new thing that is added by the wronly self-demonited “elite” it presses the question of “when is enough?”, when does all the risk and pain that would bring going against the status quo outweigh the benefits of accepting their unilateral decisions?

That’s it, its extremely obvious for anyone that is not a phishing site, except of course for Google bots, it clearly says Foofle and it doesn’t ask for email or password or any personal information, also it has unserious language like “Disunited States”.

It’s a bit weird to ask the people you are parodying to stop blocking your parody, but here we are, so if you can please spare a moment to click the link below and inform Google this is not a phishing site and maybe mention that they have a really bad AI filter that is blocking parody sites which feels a bit dystopian:

https://safebrowsing.google.com/safebrowsing/report_error/?url=https:%2F%2Fwww.forbiddensearchdetected.com%2F&hl=en-US

Thank you very much.

In my experience, this works quite well with Google’s AI, but not as reliably with ChatGPT. The main issues I’ve observed with ChatGPT are:

It has stronger and more sensitive detection systems.

It often crops images in unintended ways during the upscale process.

It occasionally hallucinates new objects or people not present in the original photo, even when the prompt is a simple “upscale this image.”

Because of this, Google’s smaller “Nano” models tend to be more consistent for this specific experiment.

Before going any further: this article is intended purely for educational and technical exploration. You should always respect the terms of service of any AI platform you use. If you decide to test any technique, you alone are responsible for understanding and complying with the rules of each tool.

The Image-Obfuscation Technique

The idea is straightforward: Apply a reversible distortion before sending the image to the AI, then undo the distortion afterward. Through trial and error, the following sequence works well:

Invert the hue of the image (only the hue channel—don’t change brightness or saturation), in Photoshop and other image edition Software this means shifting Hue by 180°.

Flip the image vertically.

Apply a reversible warp—one that doesn’t push pixels out of bounds or permanently distort details, in Photoshop the Wave warp at around 50% strength is great: it stretches pixels without losing any.

Automating the Process

Doing all of this manually in Photoshop or GIMP quickly becomes tedious. To make the experiment easier, I built a small tool:

👉 https://ivanca.github.io/bypass-image-ai-detection/

This page automates the reversible distortion step, you still need to upload the warped version to the AI for upscaling, then return the result to the tool to reverse it—but the whole loop only takes a few seconds.

For best results, use a simple and strict prompt such as:

Upscale this image, keeping the exact same aspect ratio.

Here’s a video of the entire workflow:

I know the current version has a few rough edges, and I plan to refine the UI—mainly by adding multiple distortion options for cases where the default warp isn’t giving the expected results.

ChatGPT’s own code-generation capabilities (the Codex preview) helped me build the tool used for this experiment—which adds a fun layer of irony to the whole thing.

I’m currently available for Front-End or Full-Stack development roles, including freelance opportunities. If you’re interested in working together, feel free to reach out—my email is ivanca at gmail.

That’s right, its gonna do nothing and suggest something else that it wasn’t any of the 2 initial suggestions!

Whoever team or person is on charge of the vscode autocomplete behavior at Microsoft (or at least the C# Dev Kit plugin) please do your job and fix this, thank you.