I'm not sure if it will be possible for a V3 without baked vae, since the base of it was 2.1 which also had baked vae, but i'll do some research to see if its possible to remove a baked VAE, if not for V3, then i'll try to make V3.1 have an option of non-baked.

Aparently when the parents have a baked VAE, the settings just pass through even if i select none...

The VAE that is baked in is the ' kl-f8-anime2 ', before i was using the orangemix vae but it made the colors look washed out, now in the V4 i was able to make a version without the baked-in VAE!

I found out that in invokeAI theres a way to replace the baked-in VAEs when you convert a model from ckpt or safetensor into diffuser, so thats what i've been recommending ppl do in the V3!

Thank you for offering to help!! I'll let you know if i need help with something, for now the most time consuming part is generating the images using multiple different models and then comparing them to see which results are worth keeping

Helloo, hm... make sure your stable diffusion is updated to the latest version and for the other settings:

When selecting a sd vae, you can select none since the model already have its own vae, size matters a lot when generating images and since all of those are upscaled by 2, to remake them you'll have to divide their resolution by 2 to find their original resolution.

Most of the samples were generated using clip skip 1 which you can find in settings -> Stable diffusion.

And in settings -> sampler parameters, set the Eta Noise Seed Delta to -> 31337.

The Hires.Fix is turned on for all of the images with the settings -> R-ESRGAN, 10 Hires steps and between 0,5 to 0.7 denoising.

Some of the images may require embeddings to re-create since i took the prompt from samples of other models in order to have a way to compare.