NotSoASCII | The Controversial ASCII Art LoRA

This LoRA tries to make ASCII art which is practically impossible with SD but the results turned out pretty interesting, so thought of sharing it. It is subjective if you will like it or not.

Optimal Workflow

1. This LoRA is trained on RevAnimated_v122, so try to use the same model.

2. The base txt2img image might be blurry and have less details, I recommend using high res fix if you can.

3. If you can't use high res fix, then send the base image to img2img tab and then upscale the resolution using a low denoising strength (0.3-0.45), this may do the same job as high res fix.

4. Then finally you can upscale the image to have even more texture. I recommend using 4x-UltraSharp or 4x_foolhardy_Remacri but you can also experiment if you feel like so :)

5. You can also experiment using restore faces wherever possible.

Optimal Settings and Tips

• Instance Prompt: ascii_art

• Recommended weights: 0.7, 0.8 (more recommended), 0.9

• Add "white background" in prompt if you want black texture and white background

• Add "black background" in prompt if you want white texture and black background

• If you want colorful results, don't add white background or black background in prompt and use 0.8 LoRA weight.

• If you want to convert a normal image with img2img, use 0.6-0.7 denoising strength, it can give you mixed results though, so don't expect too much. Also use higher weights.

• Adding "pink theme", "blue theme" etc in prompt makes it look interesting.

• Try avoiding "portrait" in your prompt if you don't want any humans.