NAI3LikeDiffusion
This model merge aims to reproduce the capabilities of NAI3 Diffusion. Main way to get non-generic results is to use artist tags. See preview images for some examples.
Aims to produce wide tonal range, high fidelity images without hires fix, upscaling or custom VAEs.
This is currently my favourite non-Pony model for anime and weeb stuff.
Merge recipe:
hs x 0.5 + leekai_ x 0.5 + netaArtXL_v20 - sd_xl_base_1_0.5_0_0_0_0_0_0_0_0_0_0_0_0.5_0.5_0.5_0.5_0_0_0_0_0
hs = HyperSpectral, an unreleased perturbation model. Perturbation models are a method I developed to enhance the output of model merges.
License:
Fair AI Public License 1.0-SD
Description
FAQ
Comments (10)
Very solid model !
Did you train this in Automatic1111 or ComfyUI? I've been wanting to try and merging models, but I use comfy, not sure the best way to do it.
Hey man, do you have any list of which artists are supported?
Hey, the effect of the artist tag on this model is pretty amazing.
What are the differences in results between writing “by XXX” and “[artist: XXX]”?
Great model! Just curious, is the hs perturbed model made similarly to what was discussed here?
https://www.reddit.com/r/StableDiffusion/comments/1dfuicw/perturbed_sd3_experiment/
if so how much noise have you added for this? thanks
A foolish question: how is this model for handling things like characters holding weapons or other objects? I've noticed that's a strength of NAI, but a major weakness of Pony.
hey man, thanks for the unlimited gen web page. really like your model. hope you keep the web page running for a long time.
also, which brackets are for increasing the weight?
(), {}, [] ?
is it like NAI {} or SD ()?
i like the model a lot but is posible to have the artist and char list included or the wildcards that you are using?
Say, with the recent release of Flux, are you planning to make a Flux finetune ?
This model made my jaw drop. I love it even more than CHIMERA. Thank you for this masterpiece!