Hyperstation is intended as a general purpose checkpoint with a focus on artistic composition.
Still in early development, this initial release should be considered a beta test. I hope to use whatever info I can glean from this release to make improvements in preparation for a full v1
V0.1 beta
(currently capable of high quality outputs, but requires some patience to dial in the best settings for any given scenario), some instability with hands and other details in certain scenarios. In some cases, the model has difficulty deciding on a composition during the 1st 30-40% of total steps, leading to unresolved final details.
Settings:
Steps: 32-38
CFG: 3-7
scheduler: I'm mostly using DPM++ 2m / 3m GPU, I've tested a bit with DPM2, DPMpp_2M & Euler_A, with good results depending on scenario. (for unruly hands DPM++ 3m SDE GPU with a CFG of 4.6 - 4.8 appears to be the best choice, but more testing is needed.
Contributing models:
Fabled Illusion v7 - prodajie
YamerMIX - Nihalmania - Yamer
Artaix v3 - Wildzzz
Nevolution XL/IL v2 - Necr0x0Der
Long CLIP (Distilled) L & G - Felldude
FutariumSDXLv1 (LORA) - Wildzzz
Description
FAQ
Comments (5)
Sorry to anyone using this in the generator. I don't have the buzz right now to keep it going. I'll try to open it back up once I update.
I would like to know more about how this result was achieved.
Thanks for your interest. The contributing models are listed in the description, the largest contributor is Fabled Illusion v7 (which has it's own long list of contributing models). Essentially this is a merge of many merges, with some CLIP contributions not represented in the Unet, so the actual recipe gets quite convoluted. If you have any specific questions, I'll do my best to answer.
This model generates perfect images with prompts with natural languages, it doesn't have a good adherence in NSFW prompts but I don't know if that's the focus too, but other images are too beautiful, please make more models like this because it's too beautiful, gratitude...
Hey, thanks for the feedback. This model uses several NSFW sources but it's positioned more as a general purpose / artistic model. My main model, (Fabled Illusion), is far better at NSFW while still quite artistic. I'll try to get some updates outs soon. TY.