Copyright © 2023 All Rights Reserved
4 Min
QR code technology has become incredibly familiar and is now a standard part of advertising and product experiences. In order to keep track of car parts, a Japanese company first created QR codes in 1994. But they became popular later when people started using smartphones with good cameras. Once people started using smartphones, QR codes became more popular because they could hold a lot of information, such as website links, contact details, or even pictures and videos.
Nowadays, you see QR codes everywhere! They are used for things like scanning to make payments with your phone, checking in at places, or even getting information about a product or event. They became even more popular during the COVID-19 pandemic because they were used for contact tracing.
The rapid development of AI has given QR codes a new lease on life.
A student has come up with a model for Stable Diffusion that generates images with embedded QR codes. The project was shared in a blog post by a graduate from the Chinese University of Communication, using the pseudonym ciaochaos. According to ciaochaos, a few years ago, he and a fellow student came up with a generator for "beautiful QR codes" called QRBFT.
Although the project eventually stalled, ciaochaos decided to revive the idea by taking an image generated by a neural network and applying parameters for QR code recognition. As ciaochaos writes, the idea of "embedding" QR codes into beautiful images had been brewing for a while, but suitable tools were not available until now, with the popularity of neural networks for image generation.
To create the QR code generation model, ciaochaos used Stable Diffusion. They trained the model for extension to a neural network called ControlNet with the assistance of several students and teachers from the university. After training the test model, ciaochaos used different combinations of checkpoints and LoRa (secondary models for precise parameter adjustment) to obtain images with QR codes.
To create this QR code, you can use the QuickQR Art AI generator. For that, you need a Discord account and do not need to download anything extra.
The generator offers 25 free test generations. After that, the subscription will cost $9 per month, allowing you to generate 1000 QR codes and remove the watermark from the images.
Developers often provide free generations when they fix bugs. So most likely, you won't be limited to just 25 QR codes. While we were writing the article and testing different codes, we generated around 50 of them without being asked for any money.
Use any of the free services that create a QR code. However, we still recommend making a standard QR code because overlaying art on top of the code is already a sufficient "burden" for its recognition. The simpler your code is, the more likely the AI will handle the task. Save the QR code in PNG format.
In the left panel, on the Image Generation tab, you will see several bots with the same name, pixelml-bot. You can choose any of them; they are identical and are only used to distinguish user traffic. Then, in the message field, type the command /qrart, and the following template will appear:
In the qr_code field, upload the QR code we generated earlier. The next step is to write a prompt (a description of your future image).
Feel free to experiment with the ControlNet settings in order to manipulate the results based on the given prompt, adjusting the control weights to either decrease or increase their influence.
This technique tends to yield more favorable outcomes when working with smaller QR codes that have a high level of fault tolerance.
Embrace the creative process and remember that persistence pays off when crafting your desired image through multiple attempts.
We hope you found this post enjoyable!