Sleipnir is described as the fastest and strongest horse ever to have lived. Sleipnir is comming from Old Norse Sleipnir) is derived from the Old Norse word “sleipr,” which means “slippy” or “smooth.” This seems to refer to the horse’s ability to travel not just faster than anyone, but also to any realm. Slipping through the realms like no other being is able to, even slipping out of the gates of Hel. Odin sometimes travels on him in disguise through Midgard,
This model can do:
Photorealistic
Digital Art
Anime
Comic
Vector Graphics
Line art
Stickers
can write short text
etc
The model needs no refiner!
SFW and NSFW
it's not sure what type of model it is so i have called it turbo, lighting, and a hyper model (TLH).
For generating online you can use tensor.art
Settings:
For V1.25
the max recommended resolution is 1024x1024!
For the right CFG and Sampler look in the Gallery!
Note: to get the full potential of Sleipnir FP32 in ComfyUI you need to add this Command-line Arguments: --force-fp32 important: FP32 needs more Vram and the generation is slower
For V2.7
I have attached 2 Grids one shows all possible samplers/ step count.
the other the cfg / steps
(see Training Folder)
Changelog:
V1.25 Experimental Version
This is the first Version highly experimental
the cfg range starts at 1-2/ 4 steps but i recommend 3cfg and 7 steps to start with 1024x1024 resolution.
I like also this sampler: DPM++ 2S a
made in the Gallery an grid that you can choose your sampler !.
i made also an "Instant" workflow: https://civarchive.com/models/241631/workflow-for-sleipnir-sdxl-turbo-hybrid-v2?modelVersionId=272623
V2.7
This Version is a hybrid between a turbo, lighting, and a hyper model (TLH)
it contains parts of my Valhalla Model , Ratatoskr , Furry Enhancer, Mjölnir, and parts of the hyper lora made by ByteDance
overall better quality
more working samplers
more consistent
huge CFG bandwidth
more max res (1200x1200)
working from 4 to 200+ steps
CFG between 1-10 (see Setting grid)
Disclaimer; You are responsible for your own creations
PS: will report every sick photo that is posted here that is against the law!
This disclaimer is not here without reasons. please have in mind this is like a virtual showroom and posting extreme stuff here is like 💩 on my desk in real live 😉
made by freek22
Description
First Version (Experimental Model
Please Check out the Settings in the gallery!
For the settings please look at the description or use my workflow https://civitai.com/articles/3396
Note: to get the full potential of Sleipnir in ComfyUI you need to add this Command-line Arguments: --force-fp32 but you need more Vram and the generation is slower
FAQ
Comments (11)
The name SDXL Turbo Hybrid is acceptable not like others calling that model TURBO model.
Looks promising.
Thanks for your feedback there's a reason why I say that because it's running in its own range it has ithe training data of my current best general use models inside it . the name of the model is not a coincidence because this model travel " through" the worlds and sits between the turbo model and a normal one and has a bit an unique style that none of my other model has
I completely agree about the naming. also prune please.
@glitter_fart this is not possible without destroying the quality of the model only available in fp32 for now sorry about that. When you need a pruned version you can convert it yourself but I can't guarantee for the quality of the model then!
my suggested settings for ALL turbo-(merge hybrid etc) models, 2step sample, 2 step high res, cfg:1, Eta:.5 ETA delta:2, always discard last sigma:true, (not advised:extra noise:.25 for more details but lower image clarity), milage will vary
This keeps it at a 4step process and depending on the models "turbo" state should allow for 4x upscaling(2x recommended, .5 denoise hires)
It depends I have made tests to start at 1024x1024 then you need CFG 2-3 at 5-10 steps Note not all samplers are working (see gallery) I have used DPM++ 2S a and had good results
Have you tried this settings for the model?
@freek22 have you tried mine?
@glitter_fart yes i have but I will stick with my configuration better to say I made an new workflow for the model will publish it later that day
👍 Works just fine on an 8GB RTX2070 card - even though it's a large FP32 model vs. pruned FP16 it works just fine with Automatic1111 and my 8GB card - don't let the size dissuade you from trying it out!
and for these who like an bf16 model today i released an alternative to the "big" fp32