Guide : High resolution (Patreon)

Do you...

Want to turn a small image into a large one? Want to make a low-resolution image clear? Or the images generated by ZIT are always blurry, and using the upscale model or controlnet model for enlargement always changes the content of the image, requiring repeated modifications to parameters and prompts, consuming a great deal of time and energy, yet the results may not necessarily be good.

This LORA is designed to address these issues. It only does one thing: making the image clearer without changing its content at all.

Advanced version of High-resolution for Klein4B ( ✔️ )

Advanced version of High-resolution for Klein9B ( ✔️ )

Below is a comparison with Klein.

Comparison with input

Whether it's photos, film, animation, stylized illustrations, or digital art, all can be enhanced to high definition with a single click. If you want to build your own workflow, it is recommended to use it in conjunction with the upscale model for better results. If you find it troublesome, you can join my Patreon channel, which provides an exclusive workflow (only available to members, one-click to 4K / 10K).

Precautions:

An experiment was conducted here. The original image was 256*256, divided into a 16-grid. After being restored by the model, each face was well restored, indicating that the model has strong performance. However, this does not mean it can enhance any resolution to high definition. In fact, I still recommend that the original image have a resolution of 256*256 or higher.

Image compression leads to a significant loss of information. As shown in the figure below, you can see that after restoration by the model, the details have been clearly recovered, and the restored image looks more realistic than the original one. This is incredible! In the past, artificial intelligence simulated reality, but now, artificial intelligence makes reality more real!

However, if the original image has many issues, the model will only make those issues more apparent. The following image is from MJ, with obvious hand issues.

However, if the original image has no obvious issues, such as the following image, also from MJ, where the face of the black person accounts for a very small proportion of pixels in the entire image, compression has made the face featureless, but the model has demonstrated strong restoration ability. Nevertheless, I still recommend that, to pursue precise facial restoration, please choose images with a relatively large pixel area.

During extensive testing, I used some photos from the 1980s, all of which had very low pixel counts and carried the unique vividness and blurriness of that era. The model excellently restored them, demonstrating its strong color restoration ability, which I have never seen in any previous model, and its depth-of-field processing was also quite restrained.

OK! That's all. I hope you'll like it.

Disclaimer: I holds no liability for any images produced, including their legality, ethical implications, or potential misuse. By using this model, you agree that all responsibility for generated content rests solely with you. Use responsibly and within legal boundaries.

🫡 May the AI-power be with you, see you soon !🫡