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
Important:
This is a BF16-Checkpoint so keep in mind that it ONLY works with RTX- Graphics cards.
Use 3.30 instead if you have a AMD or Intel one
FAQ
Comments (23)
What's the difference between FP16 and BF16? Other than the GPU availability.
It seems to be more precise than my FP16 version and acts more like FP32. Other than that I noticed a tiny improvement in generation speed. During generation and because i mixed in a small Lora I created there are also differences in the picture. The man with the steampunk googles is the same prompt/seed as in Version 3.30
I added a comparison from my tests in the gallery
@Afroman4peace thank you for explaining it and for showing me examples. I just haven't seen many people speaking about it that much and that's why I took the opportunity to ask you.
@ElCuajero no problem. I still need to test everything out myself. I will tell you how much speed you can actually gain.
@Afroman4peace Reading again your first response in where you compared them to FP16 & FP32, wouldn’t that mean that BF16 is like a merge of the speed of FP16 and the precision of FP32? Or am I getting this confused somehow?
@ElCuajero I just found a very interesting article about it. check it out. https://www.linkedin.com/posts/furkangozukara_what-is-the-difference-between-fp16-and-bf16-activity-7095150263602229248-Ybup
About the BF16 version: Sadly this only works with RTX-Graphics Cards. Use 3.30 instead if you have a AMD or Intel one. Here is a very interesting article about it. https://www.linkedin.com/posts/furkangozukara_what-is-the-difference-between-fp16-and-bf16-activity-7095150263602229248-Ybup
During the training did you also check the Full bf16 training (experimental?) Or just mixed and saved precision to bf16. I am guessing the first option based on model name, but still want to make sure :)
No Hephaistos wasn't trained with bf16. This is a improved version of the 3.30 version. I used this converter https://github.com/Akegarasu/sd-webui-model-converter to convert it to bf16. I just wanted to test if I would get any improvement in generation speed. In my tests I was pleasently surprised that it gained more quality. I had to bring it out as new version because it also altered the outcome compared with version 3.30. In the future I will do a full bf16 training instead of a fp16 training. This model is more like a test how bf16 works in general on other systems.
Difference is mantissa between half and full float usually.
Although I'm confused here, you say "RTX" and then no AMD or Intel. But Nvidia has cards that are not RTX cards from previous generations that handle floating point precision, did you mean you have to have an RTX model or just Nvidia?
Awesome. A must have model. I used RealityCheckXL as refiner - fantastic combination.
... I... you could use other checkpoints as refiner?
I have to test this myself;-) Thanks for letting me know about this combination.
Where did you find the model for creating a great sword like in your sample images with the cat? I can't find good weapons anywhere in the XL world.
Good question. Version 1.1 was capable to create good swords as well... I am not completely sure which addition enhanced the quality of swords. it could be SDVN6-RealArtXL or FenrisXL. Those two models are into V3.31. My additional training was more focused on faces and adding more details. I haven't trained it particularly on swords.
can anyone help with that - can I use this model with Stable diffusion 1.5
pls help
Thank you
Hi, to run this model you need at least version 1.5 of automatic 1111. I recommend to update it to the latest version though (I would recommend to do a backup in this case. Also you can install the newest version of Comfy UI or SD Next.
Is it necessary to credit the author when using this model, even when publishing the generated images?
Or is it only required when releasing a model that incorporates this amazing model?
If you merge this model into your model it would be nice to mention me in the description. If you use this model on your website for generating images it would be required to mention me. Lately I had some bad experiences in this regard. The wrote it like that it's their own model.
For the images created with it it's up to you ☺️
@Afroman4peace
Thanks for your reply!
I think it is out of the question to disguise the creator of the model in the first place!
If I ever present a merged model, I will be sure to clearly state that.
Check out the latest Hephaistos model
https://civitai.com/models/241445/hephaistos-nextgen-dpo-swfandnsfw
Why don't your Hephaistos models show on your profile? Neither on the overview nor models page.