Takes an image from text prompt and
- samples it
- latent upscales
- resamples it
- repeats
- uses a 4x upscale model to increase the size again
- saves it at a total of x8 size
Basically a series of hi-res fixes then an model upscale
Moving noodles switches between model vae and vae loader.
then image is model upscaled with lollypop x4 (not shown since 26mb image)
Changing Schedulers and Samplers (not to mention models) can dramatically change results
- Changed scheduler to noromal
- steps to 12
- Switched to regular VAE decode
- Dreamshaper model
WAS nodes
Model
Uspcale
Embeds
- EasyNegative https://civitai.com/models/7808/easynegative
- bad-hands-5 https://huggingface.co/yesyeahvh/bad-hands-5/tree/main