READ "ABOUT THIS VERSION" for GEN info -->
New V7 has both a TURBO and a LIGHTNING optimized version - superfast!
My BEST SDXL MODEL TO DATE - Short and sweet prompts are the best!
TURBO V2 - Now even better quality at lower steps!
LCM UPDATE: 1-2 seconds per generations! READ "ABOUT THIS VERSION" -->
UPDATE VAE with FP16 fix for better details: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
V.5 is up! Better everything - enjoy!
V.4 is up! Better photorealism...AGAIN!
V.3 wants DPM+ 3M SDE and V3 also has a new better license!
Image compatibility between COMFYUI and A1111 - same image everywhere! This breaks seeds and you will not be able to get same image as me without these changes! Read more here: https://github.com/Mikubill/sd-webui-controlnet/discussions/2039
↓ Settings and recommendations down below ↓
Easy and complex at the same time, this model is very versatile in the right hands. Better photorealism in XL is here.
⋅ ⊣ Why?
Look no further. The era of sharp versatile models for XL is here, in big part because of this awesome community. This is a model accumulating upon the knowledge provided by the SDXL 1.0 model and the incredible base it has given us - thanks to the team over at StabilityAI!
But as many have noted, there is always room for improvement. This model aims to take XL generations to a new plateau on which to build further and generate some really cool images along the way - be it photographs or digital art.
Realities Edge (RE) stabilizes some of the weakest spots of SDXL 1.0 base, namely details and lack of texture. Sometimes XL base produced patches of blurriness mixed with in focus parts and to add, thin people and a little bit skewed anatomy. The diversity and range of faces and ethnicities also left a lot to be desired but is a great leap forward since the days of 1.5. Lastly, the art in all its different styles and forms. SDXL base is far more capable than it's predecessors and a huge upgrade for us to play with, but there is some art styles the model still struggles with. The additions made to RE in this regard is big.
SDXL was released to all of us here. Now we build.
⋅ ⊣ What?
A methodical chaoswarp* of the best available models on Civitai combined with custom, unreleased, XL Loras I've been training these past weeks have resulted in this model. It is capable of photorealism and natural photography but that just scratches the surface. RE can do NSFW and has great anatomy information paired with Loras for better skin-texture and more realistic faces, eyes and mouths. The hole slew of anatomical corrections have been mostly fixed for the ladies and hands have also improved a lot giving way to staggering realism. The men still have room for improvements, but with this as a base I think that improvement will be here quick.
Realities Edge is first and foremost an art machine. Bombastic oil paintings, atmospheric art photography, futuristic 3D, all forms of digital art and anything in-between. If it's been expressed in art in some period of the human history, RE should be able to handle it or at least give you a great base to train your own stuff with! Loras are more accessible than ever with SDXL being the easiest plattform to train on (but a hard one to master 😉).
RE has a wide array of art-styles to choose from and most of them come out sharp and vibrant ready for further tweaking if and upscaling if needed. Illustrations, vector, oil paintings, watercolor, vintage cameras like Kodak and Ektachrome; product photography, concept art, macro, portraits, animals, comics, characters, Western-style, Eastern-style, medieval, RPGs like D&D, mechanical parts, aliens and all of these can be combined, twisted around, merged and re-synthesized into whatever concoction you can imagine.
⋅ ⊣ How?
Leaning heavily on the fantastic community model maker socalguitarist 's XL models infused with high volumes of my own acidic Loras burning off bad quality, low resolution, wonky eyes and airbrushed skin texture and adding a needed boost to creativity and range of style that the base model from StabilityAI lacks. Together with the training of the community merged within, this model shines.
There have been around 17 iterations before arriving at this one. Models have been merged with regular Checkpoint merging in both Weighted sum and Add difference but the heavy lifting was done in MBW (block merging). The many Loras were trained with Kohyaa-ss with dim rank of 256 for the sharpest detail and highest quality possible, at the cost of diskspace.
Speaking of which. Total footprint for model is ~170GB.
The model is capable of producing some basic anime, but don't despair, during the process there was an anime Lora born from the potion mixing- scheduled for release at the end of August. But that's for another post.
NO REFINER NEEDED
⋅ ⊣ Capabilities and recommendations:
Photorealism, 3D, 2.5D, Illustrations, Photomanipulation, Portraits and much more
NSFW capabilities
Works very well with Loras - both as a base to train on and for rendering
Excels at both types of CLIP prompting. Be it maximalist OpenAI style prompts or minimalist story driven LAION prompts (written in a more natural language without constant commas).
Great lighting and shines with easy short prompts and aggressive (but short) negative prompts.
Very low risk of burned generations even on higher CFG - recommend 5.5-15
Responds amazingly to hires.fix with just a scaling of 1.0-1.5 and beyond. I like doing it with no scaling at all, but just letting it run through with a sharp upscaler and less steps. If you have the VRAM for it, push the scaling higher, go nuts!
Favorite resolution ranges are 768x1344 and 1024x1296. Works good for landscapes on even bigger resolutions. Also works with anamorphic lenses in resolutions of 1920x816 or the likes. Test what works best for you.
DPM+ 3M SDE Karras recommended but always test your favorite!
All the img2img modes work really well and balancing a low CFG with a higher-than-average Denoising Strength will produce a sharp and clear upscale full of interesting details, using regular SD upscale. I wonder what you can do with Ultimate SD upscale?
Likes Clip Skip 1-4. I frequently use 2.
Knows about some celebrities - good LoRA base!
Use with ToMe (token merging) in A1111 (I'm sure it's implemented in Comfy as well) for a much faster SDXL generation time - changes seeds though!
* = The word "chaoswarp" is defined by large amounts of coffee and lots of nights spent waiting by the computer dreaming up ever more complex prompts, folding styles, tales and characters into elaborate images. In the haze of the late hours, ideas and experiments unfold that forego with such haste that having any recollection of the exact steps taken is, at this time, impossible.
Generate responsibly.
"Like ReV and RV but for XL - amazeballs!"
- some dude on the internet
⋅ ⊣ tack och på återseende ⊢ ⋅Description
New V7 model turned TURBO - Fast but not as fast as Lightning!
CFG 1-2
STEPS 4-8
DPM++ SDE or DPM++ 2M SDE
More contrast? Higher CFG or steps!
FAQ
Comments (65)
SDXL Lightning is here - Superfast and sharp images in only 4 steps!
Did you only merge the 4-step lightning checkpoint? Or did you include any of the others as well? Your opinions on which is better, lightning or turbo? Excited to try this out!
@MysticDaedra It is only the 4 step one since there is a little bit of Turbo in this as well, so creating images on 2 steps is also possible but YMMV of course. I think Lightning is better but Turbo can create cleaner faces in the distance. Although, I do recommend 5-7 steps for Turbo, so might just be the higher step count as well. I think though, that for most people, Lightning will be the new one to go with since it "out of the box" creates more punchy images whilst Turbo still needs some love in an editing program to really shine.
For that last reason I use TURBO with images I really want to control and refine while Lightning is more for fast rougher sketches - unless I am upscaling, then I like Lightning better!
@olivetty thanks, super helpful response! I guess I'll start experimenting.
@olivetty Hello, need help
1 What is the recommended sampler and scheduler and number of steps for your lightning model?
2 V7 Normal model + Choice of original Lightning LORA, sampler and scheduler?
Thank you for your work
Thank you for your continued updates and improvements. Can you give any kind of brief summary of when you'd prefer to use Turbo and when you'd prefer to use Lightning? Which for what?
Thanks so much! It is very based on personal preference I'd say since both can produce great images. I do tend to see a little bit more contrast and "spice" in the Lightning one, but if I want to control the image much I'd go with the TURBO version!
The lightning version is incredible. It can handle people of multiple ages, sizes, ethnicities and genders. It's also great for cosplay stuff, which is very rare. 10/10
So glad you like it! I think the training has some tagged characters in it so should be able to mix them pretty well! Thanks for the kind words, in a world where 3 models are dominating 90% of downloads it's great that people still "find" these smaller ones ;) <3
My lion looks with less contrast. Do I have to use a certain VAE? thanks.
It has VAE baked in, when watching the lion you can COPY the generation parameters into A1111 and it should give you the same image. Be sure to set VAE in A1111 to NONE or AUTOMATIC
@olivetty I have the exact same lion only the image is less saturated.
@sonidosenarmonia Well, it has the VAE baked but be sure to set VAE in A1111 to NONE or AUTOMATIC for it to be recognized, otherwise it'll use the one you provide yourself! :)
Hmm, maybe I did not bake VAE, just use the regular SDXL one!
Hi, Does the V7 version need VAE ? Your model is excellent. Thanks a lot
They have VAE baked! Thanks so much! <3
It says 401 Unauthorized when I use wget to download this.. ??
It seems CivitAI might've changed how their API responds to external commands. I had another person telling me the same. I have always had "Must be logged in to download" so I don't know what has changed on their end!
Good model in image quality and creativeness but it seems like it's only been trained on close ups, head & shoulders, and full length portraits. Almost impossible to get a medium wide shot.
This is somewhat true, the dataset was slightly skewed towards the former mentioned. I'll see what I can do for V8! Thanks for the feedback and good luck with your rendering! <3
yeah, but just outpaint using other models :D. This is the great thing with stable diffusion. You don't need to fit it all in one model to make it a 50000GB model that requires a nuclear factory to run. This is why I'm sticking to SDXL and ignoring stable cascade. SDXL and 1024x1024 is enough. You can always outpaint or upscale afterwards. I never really understood why people need the "entire" image in one single go.. Finetuning&Loras for 1024 is the future, no need for more.
Lightning model is good at real photography style, not friendly to other art styles (using sdxl prompt style node in COMFYUI to modify the art style can not make it draw cartoons, comics, etc.), input other art style effects in the positive prompt word input box it still produces realistic photos. The lightning model doesn't get smaller, does it mask some data during training to get faster generation? Or need special keyword activation? (Same problem with the original Lightning LORA)
That is true, but you can use LoRAs to get it to do what you want! :)
The model description says "Test CLIP SKIP 2 if your image comes out strange".
I noticed actually a trend with SDXL models that clip skip 1 turns out weird.
After a fruitful discussion on reddit, I learned that the base clip skip for SDXL is apparrently the "penultiamte layer" which means the base is apparently clip skip 2. Apparently you can override that and use clip skip 1, but it will result in weird outputs.
So I wonder if this is known in the community and whether this affects model training? That would require XL model creators training at clip skip 2 instead of 1?
Good question, I haven't really thought about that! :P
No, If the model is thrained coreectly ie.. lightning will run best based at clip skip 1. There is a possiblity that merged models will run better on C S 2. But a true lightning model will run perfectly on 1, Ive developed 2 lightning models. they can be a right bugger to lol
Whats the difference between the Lightning V7BakedVAE and Lightning 3-4 steps files?
Not clear from description
It shows on the side that baked vae uses sdxl v1.0 where lightning 3-4 steps uses lightning.
@rivdemon1221554 Thanks for chiming in!
not being able to attatch it to batchlink downloader automatic1111, it says "athentication password failed", don't know what's going on...
that would be a A1111 issue, not a checkpoint issue. go to A1111 github and inquire there.
@TheP3NGU1N Yes, also, I require the user to be logged in to download the model! :)
From Extension for A1111: Some models request login to civitai to download. So you need to set a civitai API Key on Civitai, then fill that API Key into this extension's setting page. Check document on github (Civitai API Key Section) for detail.
Don't know if the above works since I only use Comfy nowadays! <3
i can use regular realities edge sdxl with lightning lora it will equal using the lightning realities edge sdxl in terms of quality and time ?
they will be the exact same , if so same concept is true for any sdxl+lightning lora and sdxl lightning version of same ckpt ?
There will be some discrepancies just because of how lora loading effects the unet/encoders I guess but they should be pretty similar! I prefer using it like this myself just because it gives me the ability to choose longer generations if I feel like it! :)
when it comes to turbo vs lightning for any exact same model , what the differences ? which you recommend for quality and which for speed ?
Both are really fast and will give great results, there should only be some seconds in difference between the two, so both are equally fast I'd say. As for quality, I tend to like Turbo better for composition but Lightning will give sharper images, especially when paired with Dynamic Thresholding. You should try both for your specific needs and you'll find out! They are both very versatile!
Using the prompt "american girl" I tend to get plastic dolls rather than real looking people. Other ethnicities do real looking people with the same settings. Ie. "english girl", "japanese girl", "jamaican girl" etc.
Any idea why the model skews Americans into being toy dolls?
that's a weird one, although to be fair, America is just a country founded by all those other ethnicities so America doesn't really have one of their own :P JK, but I'll take a look at the tokens!
Because American girl is a brand of toy dolls as already stated.. so your asking for a doll.
@olivetty , I think you know your model is exceptional. Anyway, (among several questions, of which I'll ask a few or atleast one, to begin with) - how did you get it to follow abstract prompts? both SDXL 1.0 and Lightning...
I humbly agree, and thank you kindly! Do you have any specific abstract prompts you can share so I can get an understanding of what you mean with abstract? It can be a bit...well...abstract! :P
For one, I do come from a DiscoDiffusion background, in that AI (way back now) you had to be really creative and almost a poet to get interesting images - I have some of my learning prompts in there but not at all at scale, so shouldn't be that! :P
I've tried many SDXL models, and despite newer ones and more 'famous' ones, this model has been my go to in the past 2 months. It's simply the best for esthetics and quality, and consistently good images that other models simply cannot do. Here's looking forward to V8!
Thank you! I was actually just benchmarking it against Juggernaut X and my model is consistently better in almost every scenario, so I am really glad you feel the same way - makes my bias less biased! :D V8 it going to take time, but it'll be great! ;) <3 Thanks for the patience!
I like this model, when is the V8 planned to be released?
@olivetty I agree with you as well. Juggernaut seems really overtrained and struggles to follow prompts (despite the marketing). Yours is great -- really looking forward to what you're able to do with SD3 when that comes out
@ggyxzhoulei118498 I was recently headhunted into becoming a technical artist and AI artist at "SUPER-LARGE-AMERICAN-CORP-ICAN'T-DIVULGE" and so it'll probably slow a bit but I am still planning to get it out before July! It takes a lot of time to train these to the quality I am after! Maybe in the meantime we will get SD3? If the community wants to adopt it, that is the model I will train instead I think, but we'll see, I really like SDXL as an architecture and the quality is great!
@caverave92433 Awesome, thanks and stay tuned!
Such a great model. I think you're not getting the recognition you deserve because it's coming up as a reality model. Since you describe it as an "Art Machine" isn't that better branding for it?
Thank you so much! I feel the same actually! If you have any suggestion then please feel free to brainstorm here - I can't really change the name per-se since it is really on the "Edge of Reality" for most usecases but I am in the works creating a more artsy one as well! But again, totally agree that it is great, I've actually been using it for like months and moths at this point and it is still hitting me much better than any Jugger ever will! <3
thanks for your amazing work.
can I ask which provides better quality? turbo v7 or turboxl v2?
Turbo V7 will have the better quality!
I'm looking forward to the new model, how much longer do I have to wait? May I ask how to enhance the artistic sense of the image, how to improve one's artistic vocabulary, and if there are any recommended books or websites?
I have switched to FLUX so the wait will be a little longer I'm afraid, I need to scrap the work I already did and pivot. FLUX is just the better model and going forward I feel it will have all the controlnets and what-not so we won't miss SD that much. It is coming along nicely so have no fear - RE will be here! :D <3 The only thing I can come up with to help you with that is to read more books, be interested in art that is outside of your zone, go to galleries and museums, listen to all kinds of weird music and be BORED from time to time, let your mind wander a bit. Good luck!
Can't wait for a Flux model as this is my all time favorite SDXL checkpoint for artistic styles and textures.
Very happy to hear that, thanks! Beside some small technical issues the FLUX one is really getting there! It is far from done/perfect, but will release when I feel it is good enough and then iterate upon it!
I find myself being drawn back to this model again and again. For my purposes, its output is a bit anime-y and stylished by default, but it responds very well to negative prompts. It's much better at consistent and coherent background detail than most of its peers, and in general there's a pleasant… solidity to its rendering. It doesn't overdo detail, skin or otherwise. Its prompt adherence is at least as good as that of any other SDXL checkpoint.
Thank you for the kind words Sharlin, this one was a labor of love for sure! I am currently experimenting with getting the same "feel" with Flux but it will take a while. Your insightful feedback is greatly appreciated since it helps me condense the output from RE models into FLUX. I wish you a pleasant journey of life! :D
I've been using this model for about a month, and I can't find any flaws. Its superb adherence to prompts means I don't even need to use LoRAs. Everything I’ve tried works without additional models, which I can't say for most others. I only keep my prompts private because some take hours of tweaking to get just right. I also use Flux.1 [schnell], but again, without LoRAs.
For me, a good model is one that delivers what I need without extra calls, resulting in better quality and speed. To be honest, I get more poor images from Flux.1 [schnell] than from this model. All the images I’ve posted in the gallery have been created using just your model. If the prompt is accurate, add-ons shouldn't be necessary.
I'm eagerly awaiting your Flux version too, no matter how long it takes—it’ll be worth it. I've won several AI image awards on other platforms (I’m not too concerned with Buzz points!). I’ve also entered a few pieces in the Flux Competition here, and they seem to be doing well.
That said, my focus is still on your models—they've been solid all around. Thanks!
Awesome to hear! So glad it's working out for you. Yeah, I've been tweaking the FLUX model and am training it to be close in "feeling" to this one if that makes sense. I am not training on AI generated stuff though, don't wanna ruin FLUX, just enhance it!
whcih version did you test?
when are you releasing a Flux version :-D ?
Your SDXL model was the best XL model out there!
Because of the quality challenges faced while training a FLUX model (because it is distilled and therefore not really trainable) I've decided not to do a FLUX model. With that said, HiDREAM seems totally trainable and is high on my list! ;) <3 Thanks so much for the support and understanding!
Details
Files
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
Mirrors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7-civitai.safetensors
RealitiesEdgeXL.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
RealitiesEdgeXLLIGHTNING_TURBOV7.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.