All credit to rinrinrinrinr
This is a general purpose model able to do pretty much anything decently well from realistic to anime to backgrounds
All the images are raw outputs (with some using highresfix) no postprocessing involved or img2img upscaling
Prompt style
Keep it short and simple. If you want realism simply add (Realistic:1.5) to the beginning of your prompt and don't overcomplicate things
I recommend using DPM++ 2M Karras as the sampler
The images uploaded were upscaled through
latent(bilinear) 0.55-0.6 denoise upscaled by 1.5-2.5 OR
ESRGAN-4x 0.3-0.5 denoise upscaled by 1.5-2
Where are the negative embeddings you used?
https://huggingface.co/embed/EasyNegative/tree/main
My generations are all desaturated and not colourful like yours?
https://stable-diffusion-art.com/how-to-use-vae/
https://huggingface.co/yesyeahvh/bad-hands-5/tree/main
these are the best ones
My generations STILL don't look like yours?
Did you use clip skip 2 and xformers AND the negative embeddings?
The images with longer prompts are very old generations I made back in October of 2022 there has been ALOT of changes since then and you might have a hard time replicating those gens.
What vae did you use?
For most of the images above I used the orange vae.
https://huggingface.co/WarriorMama777/OrangeMixs/blob/main/VAEs/orangemix.vae.pt
However I would recommend https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt as It gives you much more vibrant realistic colours
My generations have ugly faces?
After some recent updates to auto1111 some things have changes. One noticeable thing is the order of the prompts you use matter ALOT more. Try reducing the prompt and use HIRESFIX or generating at a higher resolution
I want to replicate the first image in the previews?
