AI Disturbance is a feature that applies noise to illustrations to interfere with the fine-tuning of image-generating AIs (such as Stable Diffusion).
When an illustration to which AI Disturbance has been applied is used in AI training, the noise prevents the AI from correctly interpreting the original style of the illustration. This is a useful feature for those who wish to protect their unique artistic style from being imitated by AI.
Tap on ①[Back button] and choose ②[Save Artwork].
A popup window for saving the image will appear, turn on the ③[AI Disturbance] and tap ④[OK].
You can adjust the intensity of noise by moving the ⑤[Slider].
Once adjustments are done, tap on ⑥[Save].
The AI Disturbance filter has been successfully applied.
The AI Disturbance applies noise to illustrations to interfere with the fine-tuning of image-generating AIs (such as Stable Diffusion) using technologies like LoRA and DreamBooth.
*Fine-tuning refers to the technique of training an AI with a relatively small number of illustrations (roughly several dozen) to enable the generation of images with similar art styles or characters.
By applying noise to illustrations using this feature, you can degrade the quality of images produced by the AI when these illustrations are used in the fine-tuning. Below, we introduce the effects of the AI Disturbance.
First, we conducted the fine-tuning on Stable Diffusion using LoRA based on 18 illustrations by the same artist (see the collage above).
The AI to which the fine-tuning has been conducted generated the images above. It is evident that the style of these images resembles that of the 18 illustrations used in the training process.
*The prompt used was “sks style illustration of a girl with black hair, best quality, highly detailed, masterpiece, high resolution, 4k”.
Here we applied the AI Disturbance with 80% noise intensity to the illustrations.
After that, we conducted the same the fine-tuning using these images and generated the images again. Not only was the image quality degraded, but effects such as disruptions in composition can also be observed.
With noise intensities of 67% and 50%, the results are as shown in the collage above.
Thus, increasing the intensity of the noise can result in a stronger disturbance effect. Consider this as a reference when using the feature on your own illustrations.
*Note that the effectiveness of the disturbance effect depends on the AI and the fine-tuning algorithms used. This feature does not guarantee that the disturbance effect will be achieved in any situation.