This SD 1.0XL checkpoint is based on my XL3 model
https://civarchive.com/models/118312/xl3-experimental-sd10-xl-fp16
So this XL3 is a merge between the refiner-model and the base model. Also I merged that offset-lora directly into XL 3. Then I added some art into XL3. This I added a lot of details to XL3. For the eye correction I used Perfect Eyes XL. The result was good but it felt a bit restrictive. So I merged a small percentage of NSFW into the mix..
Overall it was quite more complex than it sounds.. crashed a lot during merging...
Thanks to freek22 who helped me creating this. I highly recommend his new checkpoint FenrisXL. Check out his amazing work.
https://civarchive.com/user/freek22/models
This checkpoint is intended for creating SFW and NSFW Images, art and everything you can imagine. It's not intended for porn.
CaN Tips made a really good comparison video: (leave a like there)
Many images in my showcase are without using the refiner. native 1024x1024; no upscale. It works very well on DPM++ 2SA Karras @ 70 Steps . Euler a worked also for me. Feel free to experiment with every sampler :-). Use a low value for the refiner if you want to use it at all.
Tip: Use the SD-Upscaler or Ultimate SD Upscaler instead of the refiner. This gives for me the best results ( see the example pictures).
If you use Comfy UI.. Here is my workflow for Comfy UI https://civarchive.com/articles/1648/comfy-ui-workflow-xl6-hephaistos
You also can download the VAE and use it with the standard V2.10/V1.1/V1.0. (You don't need this for the 2.1_baked_VAE version) .
Here is the link:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
Why FP16? Well it needs less VRAM to run and is better to train then FP32. This is very useful for those with less VRAM. Also this prevents an error when you want to run it in FP16 with A1111. Don't use the flag --no-half. To save even more VRAM set the flag --medvram or even --lowvram (this slows everything but alows you to render larger images).
Daedalus_7 created a really good guide regarding the best sampler for SD 1.0 XL
https://civarchive.com/articles/1501/best-sampler-for-sdxl
Flibertygibbet created a incredible guide for prompting and the use of the cfg scale with SD 1.0XL models.
https://civarchive.com/articles/1834
Thank you very much.
I merged those things into the model:
SDVN6-RealArtXL by StableDiffusionVN
https://civarchive.com/models/118114?modelVersionId=128795
first_nsfw_for_sdxl_v1 by sevenof9247
https://civarchive.com/models/118794/firstnsfwforsdxlv1
Perfect Eyes XL by Deizor
https://civarchive.com/models/118427/perfect-eyes-xl
V2.1:
freek22 helped me again training my model with some high-res pictures also some parts of his project FenrisXL made it into the model.
https://civarchive.com/models/122793/fenrisxl
Thanks for all your work.
Give me some feedback when you are pleased with the results. Also I will link my other projects down below ;-)
Also if you want to support me you can do this here.
Description
- This version got the VAE with the FP16 fix baked into it. Nothing more nothing less..
You also can download the VAE and use it with the standard V2.10. Here is the link:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
Why FP16? Well it needs less VRAM to run and is better to train then FP32. This is very useful for those with less VRAM. Also this prevents an error when you want to run it in FP16 with A1111. Don't use the flag --no-half. To save even more VRAM set the flag --medvram or even --lowvram (this slows everything but alows you to render larger images).
FAQ
Comments (10)
Thanks to the right use of my new baked VAE Version and the flag --medvram I have now new possibilities creating better art with XL 6 Hephaistos. I highly recommend to use of the costom highresfix extention. I will release a new guide for this soon :-)
In support of your march of VAE uprising, I did some fundamental research. With no negative prompt at all and only a very basic positive prompt:
VAE - Mostly black and white combinations of abstract art, female androids and occasionally a woman painted in a sea of colors
VAE Uprising - Spectacular war scenes that could be cover art of a book or a film poster, for some reason a lot of the heros depicted have angel wings
Adorable VAE - Cute babies, I wouldn't really count on support from them
Finally my favorite one
Pretty VAE - Lots of doll like pink princesses with lots of fancy jewelry (I might post a few of them later)
Hope it helps in your efforts to bring the VAE to the masses ;-)
Remember like diamonds, VAEs are forever!
@pwsoft Thank for supporting the VAE movement.. I also love the tatoo one ;-) Really good work! Your images made up my day. Do you think that Hephaistos is better than my "olde" 1.5 models? I am constantly working on it to improve my 1.0 XL models.
@Afroman4peace I was afraid you would ask me that ;-) SDXL in general has some strengths over 1.5. Closeup faces and variety in faces are strong points, text is now somewhat possible still far from predictable what you get. It is faster when running at high-resolutions, but 1.5 has much longer fine-tuning history and lots and lots of loras and embeddings available. So better as your 1.5s I would have to answer yes, no, maybe, probably likely, but not sure ;-)
What I did notice (both realistic, but also in puppetry/doll stuff) with this model (not sure if general sdxl issue or your merge. My clear lack of grounded experience with sdxl model testing is shining through here). Some of the hands, although 5 fingered and fine in general, sometimes have this "shadow" finger. It is kinda hard to explain, but were you expect to be a shadow of the hand/fingers there is an extra finger basically below the hand itself. Or the index finger splits into two separate fingers, one goes straight and the other is kinda wedged in the empty space between the normal fingers. Of course it always happens during weird prompt experiments so I don't have any clear examples to share at the moment unfortunately.
@pwsoft thanks for your honest opinion. This helps me a lot to improve the model. hands are always a problem. most of the times it works very well though. I think you can improve this with negative prompts. For my 1.5 models this was solved by adding more hand training and loras.
@Afroman4peace I found an example of the extra finger problem in one of your examples as well. The green fairy druid girl reading a book in front of a stone building (with the dear-mouse? cute little bastard! in the background). I know you tried to distract me with partial nudity, but I did pay attention to the hands (eventually) and on her left hand you see the effect. Reasonable 4 fingers (thumb is ofc invisible behind the book) and then you have this extra finger going horizontally from basically nowhere. I've seen that a few times. And yes I know hands are difficult etc, but this SDXL and should fix ALL our problems ;-) Yes? Yes!
@pwsoft well well.. you found something I have to fix within all those distractions.. I am looking into this... (I mean the hands.. So i have to create more skimpy druid ladies then - all for the science off cource :-)
@Afroman4peace Yes the sufferings we need to endure - for science. For example can you imagine having 100 generated pictures of equally pretty princesses, that you HAVE to look at in detail to select the best ones for upload. Inspecting every little detail of their beautiful jewelry - for science, of course. I guess once people realize the sufferings we as model makers have to endure, I'm sure libraries full of books will be written about it ;-) I wish you all the strength that may be required. Stay strong!
I just published my new checkpoint XL13 Asmodeus.
Attention: Do NOT click on any links in the comments. (Except my links of course). I try to get rid of the those spammers as soon as possible but I can't monitor this site 24/7













