About: Trained on selected dataset with original resolution from psx esque.
For Foocus, disable all styles and negatives ( you can also try noise and low quality at negative, but it will make look like upscaled. )
This lora was trained with low resolution, so some artifacts like multiple persons will happen.
TO DO: A new version with fixed resolution and noises.
❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️❤️
Do and tips: 1 weight for generation / 1.2/1.5 weight for img2img
Keyword: low poly , playstation 1 still
tips or requests:
https://ko-fi.com/luisapinguin
Description
This is the first version of this lora.
FAQ
Comments (4)
After PSX style trend on social media; this is what I needed. Always good and interesting models.
For img2img, what are the values for denoising strenght you recommend? To get the result as the example pic you put for img2img
I'm not OP but maybe this can help!
I got best results with denoising at 0.3-0.4.
Extra note: img2img worked best in tandem with the img also in a controlnet unit in my experience so that it preserves more of the original.
Below are my settings. But note that lots of tweaking and multiple runs are required to get passable results. For my results, I needed to combine multiple results in photoshop to get something I was happy with.
img2img settings for luisap psx lora:
Steps: 24, Sampler: DPM++ 2M Karras, CFG scale: 8, Size: 512x512, Model hash: d8fd60692a, Model: sdxl-helloworld-xl, Denoising strength: 0.3, Lora hashes: "sdxl-psx: 97857fb508e5", Version: v1.7.0
ControlNet Unit 0 model: diffusers_xl_canny_full, pixel perfect: true, upload independent control image: true (same as img2img), control weight 0.9-1, control mode: controlnet priority
Are you planning on making one for sd 1.5. I have only 6gbvram and cant use SDXL. *fake tears running down mie face but none the less based on a true story. The model/lora looks so good.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.