EDIT: since V20 I created a discord community, feel free to join us, for help me improve it
Here come the realistic version of my merges, will fill out more infos soon.
All my example are done with comfyui, check about this version for settings.
VAE Included
You can find my workflows here: https://civarchive.com/articles/4338
You may also join my discord server to discuss models and share generation.
Please, share your result, don't hesitate to comment with question, Will do my best to reply.
Hope you like it...
Description
Here is the merge:
Models:
Mklan22.0: https://civitai.com/models/305648?modelVersionId=445282
Mklan21.0: https://civitai.com/models/305648?modelVersionId=401911
70% Mklan / 30% :
https://civitai.com/models/193834/realism-from-hades-xl?modelVersionId=434428
https://civitai.com/models/471585/clarity-xl?modelVersionId=524640
https://civitai.com/models/119229/zavychromaxl?modelVersionId=490254
https://civitai.com/models/443268/interdiffusion-35?modelVersionId=493656
thanks to the creators of those models !
Check samples for settings, but feel free to try more, it works on lots of sampler/scheduler as well as steps from 15 to 50.
FAQ
Comments (8)
I have tried to use this new model, but unfortunately the images are garbled, where can I find the dpmpp_sde_sgm_uniform sampler I do not have it..
I used comfyui, but you may also use forge, activate some add-ons, come over discord we will show you ;)
Is it correct that it wll give garbled images with other schedulers? I am using online service of mage.space and there we dont have that scheduler. We have DPM SDE but with that i get the garbled results. Which i dont get from your previous models. Is vae baked in btw? Thanks a lot for all your work.
I'm rendering without any problems using a CFG scale of 4 and sampler DPM 2a, 50 steps, at 960x1280 resolution.
Surprising detail for a Hyper checkpoint, nice!
Thanks
Nice results but I'm confused about 'Hyper'. Isn't the idea of Hyper supposed to be using few steps? All the examples are seem to be 30 steps which doesn't particularly speedy.
Well, Hyper works both ways, it does nice result around 6 to 10 steps, but even greater result at 20 to 30 ...this model works both ways...