I prohibit companies like PixAI, Tensor, etc. and their users from uploading my checkpoints to websites outside of Civitai. I will do it myself if I want!
☕ Buy me coffee, if you want to support me: https://ko-fi.com/tzigoai
Please rate, comment and post your results with my model! Thanks!
Sampler: DPM++ 2M Karras, but it also produces good results with other samplers.
Steps: 30 and more if necessary
If you get noise/blur increase the step-count by five to ten.
CFG: 7-10
Version 0.81_VAE
In retrospect, version 0.78 became a bit too semi-realistic for me.
This has been changed again with this version. There are slight improvements to the lighting and depth of field.
The basis is version 0.6C, which has been enhanced with Copax TimeLessXL and Copax photoXL.
Credits to Copax.
Version 0.78_VAE
In version 0.78 I added the great ZavyChromaXL and WildCardX-XL-Fusion OG, as well as some Loras.
Credits to the respective authors.
Many thanks to Zavy and Mr_fries1111 for allowing me to use their wonderful checkpoints for this model.
Version 0.6C_VAE_HYPER
Its Version 0.6C_VAE, but HYPER.
Much faster, at the cost of slightly reduced quality.
Sampler: Euler or Euler A (with Euler A completely different images can emerge!)
Max steps: 12
CFG 8
Version 0.6C_VAE
As I was not very happy with version 0.4A and I am now allowed to use the great epiCRealism XL thanks to epinikion approval (thanks again for that), I decided to put together a new version of TzigoRealMixXL.
I hope you have fun testing and I'm excited to see your results.Based on merges I already have and new attempts, version 0.6C_VAE was created after days of testing).
Merge A: Imaginarium - STOIQO NewRealityXL 0,45 - 0,55
Merge B: Real_WoW_XL - Realism from HaDeS XL 0,55 - 0,45
Merge C: Photonium - PhotoPedia XL 0,4 - 0,6
Merge D: CyberRealistic XL - C 0,65 - 0,35
Merge E: epiCRealism XL - A 0,6 - 0,4
Merge F: Juggernaut X - B 0,65 - 0,35
Merge G: D - E 0,5 - 0,5
Merge H: G - F 0,5 - 0,5 + sdxl_vae
Add the desired Loras.
Credits to the respective authors.
------------------------------------------------------------------------------------------------------
Merge recipe Version 0.4A:
As a base: Juggernaut XL - LEOSAM's HelloWorld XL 0.5 - 0.5
Lower merges:
Merge A: AlbedoBase XL - ThinkDiffusionXL 0.45 - 0.55
Merge B: DreamShaper XL 1.0 - Raelestia XL 0.6 - 0.4
Merge C: A - B 0.55 - 0.45
Merge D: Basis - C 0.55 - 045
Add the desired Loras.
Stir vigorously.
And TzigoRealMixXL is ready!
Ps.: I'd rather not even mention the hours of testing in between.
Credits to the respective authors.
Description
Merge recipe:
As a base: Juggernaut XL - LEOSAM's HelloWorld XL 0.5 - 0.5
Lower merges:
Merge A: AlbedoBase XL - ThinkDiffusionXL 0.45 - 0.55
Merge B: DreamShaper XL 1.0 - Raelestia XL 0.6 - 0.4
Merge C: A - B 0.55 - 0.45
Merge D: Basis - C 0.55 - 045
Add the desired Loras.
Stir vigorously.
And TzigoRealMix is ready!
Ps.: I'd rather not even mention the hours of testing in between.
Credits to the respective authors.
FAQ
Comments (21)
Will be running tests on this in a few, including lightning. Description is just that it's "real" though so was wondering if there was anything in particular it was aimed at doing well that I should explore?
The real is just short for realistic, although it can also generate anime images if you really want it to.
I am looking forward to your results
@Tzigo I'll make sure to tag you on them then, though the download is particularly slow tonight (sub 1mbit/s) so will probably take a while. And yeah, I was familiar with the use of real there; I phrased that weirdly, my apologies.
@CivitFowl There's no need for an apology
Right, first post made. Will explore it a bit more, make a few more posts, then write up a review.
I feel like I should ask though, how did you reach the 40 steps conclusion? I'm doing fine with far less.
@CivitFowl Thank you for your work, the pictures turned out great. The 40 steps are just a recommendation, as I did most of my tests with them. You can expect usable results even with ~20 steps.
@Tzigo Got more coming, gotta check landscapes and some other things after all.
And yeah, I saw some decent results all the way down at 10 when doing it right, but it becomes a bit hit/miss there, and slightly blurry.
Can be done but it isn't a lightning model after all :)
@CivitFowl I'll upload a Hyper version this evening, but I'm still testing it at the moment.
@Tzigo Hah, I wasn't actually pushing for one to be clear but good to hear. I'll try to start landscapes sometime this evening (in ten hours or so).
@CivitFowl I thank you for your commitment ❤️
@Tzigo Well this is how I relax honestly, been a long week so joining in and helping with your project seemed like a fun thing to do :)
Now I've started on landscapes and played around a bit more with CFG, steps, and similar but I keep getting this slight grainy effect and I'm not sure what's causing it since I've even tried the exact same settings are yours. Any idea?
Alright, review done at 130 images. Positive, obviously. Might drop some more materials in here later but other than that I think it's done until there's an update or I get an idea. :)
@CivitFowl Which sampler did you use?
@Tzigo Recommended one, turned out that going lower on steps, or clip skip both of which I did, increased the level of grain depending on the prompt. Not sure how that keeps happening to models but it's not too unusual, for there to be things like that.
Very realistic , but seem hard to generate NSFW when putting dress on , I need to use "nude" to create clearer nipple + pussy pics
That's what I'm thinking, since it's not a NSFW model. But as you yourself noted, I can't completely prevent NSFW use, since this option was available in the original models.
@Tzigo will you intend to create pony version as well? :)
@leepeter1231 Maybe, but not as a realistic model!
a NSFW version would be great eventually. Out of all the "realism" checkpoints out there, this one so far has been the best at following the prompt.
@MysticDaedra Since I'm absolutely not a fan of NSFW art, there won't be a version in that direction!