No upscale needed for high detail.
Version 4 is a much more versatile model compared to v2 and now it doesn't automatically default to realism when prompting for people. But is still very good at it.
Version 3 scenic edition comes with a few changes. With a focus on scenery, backgrounds and locations, this isn't so much a replacement for version 2 as it is an alternative version, However in general I prefer the output of Scenic Edition over V2, so give it a try and let me know what you think!
The model can also be used on lower resolutions, like 768x768 and produce good results.
Please, if you like what I'm doing, support me down below!
https://www.buymeacoffee.com/tripleheadg
Buy me that Pizza!
Models merged:
https://civarchive.com/models/175094?modelVersionId=208092
https://civarchive.com/models/147720?modelVersionId=213982
https://civarchive.com/models/155977/demon-core-sfwandnsfw?modelVersionId=177633
https://civarchive.com/models/156061
https://civarchive.com/models/144952?modelVersionId=207945
https://civarchive.com/models/193834/realism-from-hades-sdxl?modelVersionId=217874
LoveXL v2, WoW_XL v2, AnimeGodXL, 7 of 9 v3 and v2 were all included in the merge as well.
If you don't include previous 7 of 9 versions, that makes a total of 9 models merged for this update!
Description
FAQ
Comments (35)
这大小和介绍.XL了，1.0标对，2.0还能标错的
Thank you for pointing that out!
If it is the most realistic model available, then why do none of the example images look photo real? You just have a lot of Pixar looking stuff. I have seen other SDXL model that look like photos, so I think you claim is a bit out there.
If you're talking about my images, you must have left your glasses at home.
Also where did I claim that my images look like photographs? I said they look more realistic than any other model. Not realism themed, realistic. Not photography styled. Realistic. As in actual real people not air brushed magazine cover models.
My model has the best skin texture, best eye detail and best anatomical consistency of ANY model you will use. And if you can prove me otherwise, I'll go ahead and review that model 5 stars myself, before using it in my next merge.
As it stands however, I do not believe you can do that.
You're also talking to the guy who released an entire suite of photography based styles with multiple thousands of downloads.
I would like to add that the example images do NOT represent the variety and potential of this checkpoint. You can find more really good images created with this one on the Ai Revolution Discord - just stop by and have a look^^.
@Triple_Headed_Monkey I think you are confused about what realistic means. It means it looks real. None of your examples look real. They all look like fine Pixar animation. The skin doesn't look real, the eyes don't look real, therefore, not realistic. If your model can actually do realistic you should post some of those images so people can see it. If you can even get one that looks real I might download the model and try it, but I am not looking to make cartoons.
@Triple_Headed_Monkey Are any of the photography based styles on Citivai? Everything of yours I can find on here look like cartoons.
@Triple_Headed_Monkey Juggernaut XL, AcornXL, Copax all look considerably more real.
@Triple_Headed_Monkey https://civitai.com/images/2639276?period=AllTime&periodMode=published&sort=Most+Collected&view=categories&modelVersionId=159838&modelId=112013&postId=623570
Here is what realistic looks like
@AlBat I don't mean to make you look silly, but you can clearly see how many negative keywords that this image required to produce that level of realism. My model does not require this at all. In fact because of this you have much more freedom to prompt whatever you actually do want. No need for crazy negatives or triggerwords which may impact the intended composition in unplanned ways.
@AlBat Just to prove a point I am now generating images using the same positive prompt as the image you linked, without using any negative prompt. I shall provide you with a link shortly.
I will also include some images using a simplified version of the prompt without any realism triggerwords present for comparison.
@AlBat u need to use the comfy workflow as mentioned in the description then the pics a very realistic some even like photos!
Sorry I got sidetracked helping a few people over on the AI revolution discord server. Above is the link where you can see the images generated by this model.
@Triple_Headed_Monkey I'll admit they look a lot better than any of your examples do, but still not as good as the ones I did. Maybe you need some negative prompts.
@sikasolutionsworldwide709 I don't have Comfy. I can't get it to work on my computer.
@AlBat It should still work on A1111 but the process just takes longer and has more steps. In ComfyUI I press generate and it:
Creates base image > Upscales 2x > IMG2IMG > produces final render.
Is there any reason why you can't get comfyui to work? I don't mind assisting you if you pop onto the AI revolution Discord Server.
https://discord.gg/JxkXxee3B8
I've been troubleshooting issues for people on both Comfy and A1111 over there for a few months now.
@Triple_Headed_Monkey I have tried again and finally got an installation of Comfy working. Look, I know we have had some disagreement in this comment thread, and I am not just here to try to trash your model. It seems to be a pretty good model. I get tired though of the claims that are never backed up with results. This model as well as the FoddaXL model both claim they are the best and most realistic model in existence. Neither give as real looking results as AcornXL has and anyone who looks at the galleries of the models can plainly see that. Seeker70 doesn't even make claims to be one of the best, in fact says it's not perfect, but pretty good. I still appreciate your model and will try it out more to see if I can get it to perform as well as my favorites.
@AlBat That's fair enough. I've just used a new merging technique to tune the model to focus specifically on realism with simplified prompting.
I only claim anything here, because I am confident in the power of this new methodology and am hoping to bring more mergers into the fold.
Nice model, going to try it out! Thanks!
Why do 99% of the images in the Gallery have no prompt information, some guidance would've been really appreciated!
Apologies. It's a twofold thing. First of all I use ComfyUI, not automatic1111. And CIVITAI aren't bothering to work that stuff out.
Secondly I keep forgetting to set up the single available option for saving metadata in a compatible civitai format. It requires me to edit my existing workflow and I keep using it as is instead xD
I should also mention that my model's output is different to some others in that changing CFG, for example, does not significantly impact the image results and can be used to generate variations of the same images at different lighting/exposure levels.
However I can provide you with the details on any of the images if you would like.
To make things easier, hop onto discord and join the AI revolution server.
https://discord.gg/JxkXxee3B8
I've been troubleshooting issues for people on both Comfy and A1111 over there for a few months now and I frequently share my prompt information there.
I actually am one of those people who believe in being open and transparent, so I will make sure from now onward to provide as much information as possible.
I should probably mention also, as a general piece of advice, that you should treat this model as if you were prompting midjourney. Natural language descriptions works quite well, as well as using additional keywords alongside it to bolster the description.
Many of these images used single sentence prompts like:
A beautiful woman with intricately detailed pupils
Combined with the photography style keywords from my style node pack to specify camera, depth of focus, backlighting, contrast and sunset.
@Triple_Headed_Monkey Thanks for the information, will join the Discord!
An excellent model that allows you to achieve maximum detail on pure generation without using additional tools, such as "detailers" or LORA, with a proper upscale, the result is multiplied. Thank you for your work and for sharing your work and experience with users. With respect and wishes of success in further developments.
No problem! If we all move forwards together, progress happens faster. So we all win!
@Triple_Headed_Monkey I agree, progress will move forward faster if we join forces!
This is a great model in COMFYUI
fantastic results, totally love the creative aspect of it
That being said, and I am not being sure if the creator is simply unaware of it, in Automatic1111(1.6.0) this model causes pages of bugs if it's the model loaded when starting the UI.
It works kinda fine when you booted A1111 with another model, that being said, it messes with your model folders, Loras go missing, TI's disappear completely, inpainting, controlnet etc... all bug out...
reproduced the issues on 3 different PC's before posting this. The Creator should issue a warning that there is a possibility for A1111 user to run into a heap of bugs. I have one machine where it completely works fine with version 1.5.0
Other than that... amazing model in Comfy
Sorry to hear that. Hopefully the issue may not occur with the latest version.
Sorry to hear that. Hopefully the issue may not occur with the latest version.
@Triple_Headed_Monkey I'll check and get back to you. It's not a biggie at when you know about it, the problem is, when things disappear and you don't know the cause, restored version won't work... people might be starting to reinstall. :-D
@Triple_Headed_Monkey V3 Scenic Edition works just fine :-D
@DeVicious Awesome! Hope you enjoy. :D
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.