Realisian XL
Hey everyone! πβ¨
Just released the first SDXL version of Realisian - a port from the SD 1.5 version. Since it's in early access, your feedback is super valuable! π¨
Let me know what works, what doesn't, and what you'd like to see improved! π«
Feel free to reach out on Discord or X (Twitter) if you have any question! π€
Thanks for your support! ππβ§ο½₯οΎ ΰΌΊ Recommended Settings ΰΌ» ο½₯οΎβ§
β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β
β§ Sampler
DPM2 a
Note: Please do NOT use Euler a.
β§ Steps
15 (12 β 30)
β§ CFG Scale
1.5 (1 β 2)
Note: Higher CFG values require more sampling steps.
β§ VAE
None
β’:β’:β’:β’:β’:β’:β’ β’:β’:β’:β’:β’:β’:β’: :β’:β’:β’:β’:β’:β’:β’ β’:β’:β’:β’:β’:β’:β’
β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β β
β§ Discord β’ Cisney Gassai#5108
β§ X (Twitter) β’ @Realisian
β§ Ko-Fi (Support Me π) β’ ko-fi.com/cisneygassai
Description
Initial release
FAQ
Comments (21)
Downloaded, will be testing...
It's marked as Pony, though - is it Pony or SDXL?
Nice, feedback is appreciated :).
Yes, it is a Pony model, but Pony is also based on SDXL, that's why I refer to it as an 'SDXL version'.
@Cisney_GassaiΒ cool so its pony not sdxl got it.
I am failing to see any "Realism" here. Shame.
this is much better
well its not realism, its realisian
i'd rename this to Semi-Realistic as that's what it is. it's not fully realistic
The previews all display strong "sameface" - perhaps make some new ones to show facial diversity?
This model wasn't trained with facial diversity as a primary focus, though it can be achieved through detailed prompts or by using a LoRA for specific characters (since it's based on Pony).
You know what people say: "all Asians look the same" - that's kind of the issue here. Even though it wasn't trained with one single character, they all look similar.
Thanks for your feedback, it may be something I can improve in a new version.
Thank you so much for the new version!
Wish you included promts on the sample images.
Look at the images on my profile, only the images on the model showcase doesn't have the metadata, but every other image does.
the skin tone is very pale and resembles a bit 'plastic', the lighting is very dark for 'pony environments', however I am referring to your exhibits here, but I am inclined to download it.
Yeah you are completly right and that's on purpose, but I'm thinking on adding a bit of "noise" for a second version so it looks more realistic. Thanks for the feedback! <3
@Cisney_GassaiΒ I downloaded your checkpoint after posting here and made some renders..., I guess it can't do without a hires fix? Maybe you should design explicitly for FORGE because I tested it with A1111, but I'll use it with FORGE today too!
@ArtistryUnleashedΒ I use A1111, don't know if Forge needs any extra tweaks, try with the recommended settings because I don't use hi-res fix and I don't recommend to use it either. The only extra thing you may want to use is Adetailer to improve the eyes.
Can anyone please tell me how do you write such complex prompts???
u can use chatgpt for that or idk what they use but i use chatgpt (btw nsfw support chatgpt)
It doesn't seem to handle dark skin very well. I like to make black characters, because there is a glut of white women out there, and it seems the default is white skin or East-asian appearance. Any tips for black women, indian women, brown arabic women, or brown/black hispanic women?
Try adding something like this with your prompt:
(dark skin:1.3), (dark-skinned female:1.3), (very dark skin:1.3)
Thanks for the great model! It'd be helpful to have recommended latent image sizes in the description.
Thanks, 1280x1600 is what I normally use.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.