How to Create Undetectable AI Images?

[RichTapestry]

Want to make AI-generated images harder to detect? Here's a simple method:

1. Generate the AI imageusing your favorite tool:
   • MidJourney
   • Bing AI Image Generator
   • DALL·E with ChatGPT
   • Ideogram
   • Canva
   • Adobe
   • Deep AI
   • Grok
2. Print the image on paper.
3. Scan it back into your computer using a scanner or simply take a photo of it with your mobile camera.

This technique alter the pixel arrangement slightly, making it more challenging for detection algorithms. While some humans might still spot common AI errors, this process makes it harder—especially if you also resize the image to a smaller resolution.

Do you have more tips or tricks?

 

[RichTapestry]

Additional Suggestions to Enhance Undetectability:

• Post-Editing: Use tools like Photoshop or GIMP to adjust colors, add textures, or modify features to further mask AI traits.
• Filters: Apply analog-style filters like grain, blur, or noise to simulate imperfections.
• Upscaling/Downscaling: Resizing the image (both up and down) can disrupt patterns associated with AI generation.
• Partial Cropping: Cropping the image slightly can also remove recognizable patterns from the edges.
• Hand Edits: Manually redraw small details, particularly areas with common AI errors (e.g., hands, reflections, or text).

 

[Perry]

Why this works... This answer is based on research I did using ChatGPT:

Printing and rescanning (or photographing) an AI-generated image alters its digital characteristics in subtle but meaningful ways. Here's how and why this method works:

1. Changes in Pixel Arrangement:
   When an image is printed, its precise digital pixel structure is converted into a physical representation. Rescanning or photographing it reinterprets this physical image back into digital form, introducing slight shifts in pixel alignment, RGB values, and overall structure. These changes disrupt the original digital fingerprint left by AI generation.
2. Introduction of Analog Artifacts:
   During the printing and rescanning process, various imperfections are introduced, such as:
   • Lighting variations (from scanner light or ambient conditions)
   • Resolution differences (depending on the printer and scanner quality)
   • Lens distortions (if using a camera)
   • Paper texture effects (depending on the print material)
   These analog artifacts modify the image in ways that mask the AI-specific traits embedded in the original file.
3. Masking AI Characteristics:
   Detection algorithms rely on identifying consistent patterns, such as pixel alignment, specific textures, or color gradients typical of AI models. The noise and imperfections added through the analog process break these patterns, making it much harder for algorithms to detect or trace the image back to its AI-generated origin.

 

[Ant]

That's a clever approach, and it's great to see the technical details behind why it works! Another angle to consider is integrating these images into real-world contexts. For instance, using the AI-generated image as part of a physical collage or artwork can further blur the lines between AI and human creation. By embedding the image within a larger, tangible piece, you not only add analog artifacts but also create a narrative or context that can distract from AI-specific traits. Have you tried experimenting with such methods? It might be a fun way to explore the boundaries of detection even further!