V10 is sponsored significantly by Wizz (https://civarchive.com/user/_Wizz_) - check him out, amazingly kind and knowledgeable guy.
Please consider joining my Patreon so I can keep most of my work available for everyone - I'll be releasing early access models there for a cheaper monthly fee than buying them here individually, and also will be providing exclusive models and more!
For business inquiries, commercial licensing, custom models, and consultations, please get in touch at [email protected] or [email protected]. You can also contact me here through CivitAI DM or join my Discord.
If you see this model on a commercial product/website that is not CivitAI, then the website you are seeing it on is stealing and profiting on my work. Please contact me. If you currently offer it commercially, contact me to avoid being called out publicly.
Note: commercial licensing is a requirement for all generative services other than CivitAI. It's also a requirement for companies using it in a commercial setting; e.g. marketing materials, fashion shoots, apps for photo/video editing, etcetera. The only commercial usage that does not need a commercial license is for artists selling their imagery as prints and such.
Consider using DPM++ 3M SDE Exponential or DPM++ SDE Karras for this model, I personally like them best, 3M SDE Expo for most styles, DPM++ SDE Karras for photorealistic. Make sure to read the prompt advice further on in this section if you want to create photorealistic images as well as some of the sample images.
Introduction
A model line that should be a continuance of the ZavyMix SD1.5 model for SDXL. The primary focus is to get a similar feeling in style and uniqueness that model had, where it's good at merging magic with realism, really merging them together seamlessly. Of course with the evolution to SDXL this model should have better quality and coherance for a lot of things, including the eyes and teeth than the SD1.5 models. This model has no need to use the refiner for great results, in fact it usually is preferable to not use the refiner. Recommended to use ultimate SD upscaler to get the most amazing results.
Continue reading for more information and some tips and tricks.
I kindly request that you share your creations both here and on my Discord server, as I would greatly appreciate the opportunity to see them and motivate me to spend more time in further ventures here.
Pros and cons
Pros
Much better saturation than base model.
Much better teeth, eyes, hands and feet.
Increased realism.
Less blurry edges but still keeps a certain pleasing softness to the image.
Better looking people.
Great texture and tonality.
Cons
NSFW much better than base, but still somewhat lacking without LORAs.
Roadmap
Training the SDXL model continuously.
Pioneering uncharted LORA subjects (withholding specifics to prevent preemption).
Tips
To better understand the preferences of the model, individuals are encouraged to utilise the provided prompts as a foundation and then customise, modify, or expand upon them according to their desired objectives.
Ditch the refiner, an img2img ultimate SD upscaler gets better results when you select this model for it.
For photorealism, use nmkdSiaxCX_200k for initial upscale, consider using ultramix_balanced for final upscale/pass to lessen grainy pictures.
Consider using face restoration techniques when the result for the face is subpar, but the rest of the image is interesting.
If you find that the details in your work are lacking, consider using wowifier if you’re unable to fix it with prompt alone. wowifier or similar tools can enhance and enrich the level of detail, resulting in a more compelling output.
ComfyUI is the UI I use for my SDXL images.
Your 1.5 LORAs won't work in SDXL.
Consider finding new prompts, don't use the standard 1.5 ones. SDXL likes a combination of a natural sentence with some keywords added behind.
To maintain optimal results and avoid excessive duplication of subjects, limit the generated image size to a maximum of 1024x1024 pixels or 640x1536 (or vice versa). If you require higher resolutions, it is recommended to utilise the Hires fix, followed by the img2img upscale technique, with particular emphasis on the controlnet tile upscale method. This approach will help you achieve superior results when aiming for higher resolution outputs. However, as this workflow doesn't work with SDXL yet, you may want to use an SD1.5 model for the img2img step.
Prompts
Recommended positive prompts for specifically photorealism: 2000s vintage RAW photo, photorealistic, film grain, candid camera, color graded cinematic, eye catchlights, atmospheric lighting, macro shot, skin pores, imperfections, natural, shallow dof, or other photography related tokens.
Recommended negative prompts: As few negative prompts as you can, only use it when it does something you do not want, like watermarks. Consider using high contrast, oily skin, plastic skin if the skin is too contrasting or too oily/plastic. Also make sure to add anime to negative prompt if you want better photorealism, and more mature looking characters.
You are further encouraged to include additional specific details regarding the desired output. This should involve specifying the preferred style, camera angle, lighting techniques, poses, color schemes, and other relevant factors.
Recommended settings
sdxl_vae.safetensors (baked in).
DPM++ 3M SDE Exponential, DPM++ SDE Karras, DPM++ 2M SDE Karras, DPM++ 2M Karras, Euler A
Steps 20~40 (lower range for DPM, higher range for Euler).
Hires upscaler: nmkdSiaxCX_200k, UltraMix_Balanced.
Hires upscale: Whatever maximum your GPU is capable of, but preferably between 1.5x~2x.
CFG scale 4-10 (preferably somewhere around cfg 6-7)
Lightning LoRA specific settings:
Euler sampler with SGM Uniform as Scheduler.
Steps 4 (use the 4 steps LoRA)
CFG scale 1-2 (CFG 1 at the higher weights for the LoRA)
LoRA weight 0.6-1
Model recipe
Trained from SDXL 1.0.
Social media
You are welcome to join my recently created Discord server, where we can engage in discussions, share our experiences in AI, and showcase the things we’ve made with AI. You are encouraged to join and ask any questions or seek additional tips and tricks related to my models or AI in general. Your participation would be greatly appreciated.
Description
Moved somewhere halfway between v6 and v7 with regards to contrast, sharpness and darkness based on feedback.
Should be much more compatible with most LoRAs again, which was one of the main points of this model. Whereas v7 went backwards on this.
Less color banding (posterization) in color gradients.
FAQ
Comments (82)
Yeaah, lets celebrate v8 with an image creation frenzy. :)
Ahhhhhhh! V8!! I'm excited! Nefarious purposes coming right up.😈
One of my top 4 SD XL models, and it keeps getting better!
v7 was pretty tight actually and this promises to be even better! So excited!
I have to say...
I really liked v6. I held off on saying anything about v7.
v6 was bright and colorful. I was excited to try v7. But, it was a somewhat darker. I was a little disappointed. I still pumped out a lot of images with it. Overall, it was still a good model.
So far, v8 seems to be in between both. But, more vibrant on the colors. Which is a good thing.
AND, I still prefer ZavyChromaXL over any other model. And I have extensively tested the same prompt with the same seed and settings on a bunch of other models. Nothing comes close to the look that "I" am going for. It may cost more buzz than some of the others to use. But, to me, worth it.
So cheers to a job very well done on all your models! Your LoRas rock as well.
Thank you for a beautiful model. I didn't use it yet but I can tell its gonna be a keeper. Hope I can make beautiful art like you. I realize its not the model, but the person using the model, that is goal.
loved v7, can't wait to experiment with v8 ♥
thank you for your hard work!
Fantastic model! Well done!
Well, this is exciting.
Most beautiful checkpoint on civitai. Thank you for hard work sir
the faces and body are really good but cant make the hooves,genitals are bad :/ cfg over 4-5 makes the picture bad and deformed somehow
It should be able to do hooves, but won't be as good as Pony at it obviously. It's not a focus of this model to primarily have it generate nsfw. It definitely has more knowledge about it than base sdxl, but it's only intended to have somewhat of an understanding so LoRAs can latch onto the concept better. If you really want to enhance it you'll need to use one of those many LoRAs out there or inpaint with another model intended for specifically nsfw. The examples posted are cfg 6, it doesn't deform or become bad there either.
@Zavy im mostly using your fantasy checkpoint it is the best,i highly appreciate your work :D,perfect hooves,perfect with most of the loras, just not rly realistic ofc.and the faces could be better. (there are no loras for hooves or 3 breasts tho) could you please make me an example picture with this new checkpoint? i couldnt manage to make it work :(
I think this is my favorite version. 7 and 8 started looking weird.
This is V8?
Yeah. I tried 8. 6 is still the best for me.
@Silmas It is but the comments are for all versions. I think they're talking about 6.
@lokithorodin Now I understand, you have no reference in your original post. Thank you for clarifying.
@Silmas Not my post. Just being helpful. Have fun!
When I first got on Civitai I chose juggernaut because it was so highly rated. But now that I've tried a slew of checkpoints I always come back to Zavychromal. It always manages to produce a sense of magical realism I can't get anywhere else.
juggernaut is overrated & mediocre. it's only got lot of downloads/ratings because it's decent at photo realistic porn
Thank you! And yeah I generally agree with Melynn, it can be argued Juggernaut was good up til about version 7, but others have surpassed it and it's more coasting based on its prior stats than anything else.
love this 8.0 model. Great work!
Thank you!
incredibly good, simply the best sdxl model on civitai. versatile, accurate, & beautiful. produces high quality images.
Thank you, very nice compliment to read!
love the starry night picture with the James R. Eads Satori Canton, and Van Gogh combo. do you have a list of artists that work with the model or reccomendations?
Generally just the same as SDXL, but may deviate slightly more to a more detailed image than the actual style: https://rikkar69.github.io/SDXL-artist-study/
I remember checking this out many versions ago and not being super impressed with it, but I'm glad I've circled back to it. It's improved greatly in the last few versions in the way it handles ages, general body shapes, and how actual real people look. In a world where there are a billion checkpoints that can create 18 year old stick figures, it's good to have models like this around that can do much, much more than that.
Thank you, I certainly try to keep it more inclusive, but with averages being what they are that remains a struggle to get somewhat diverging results. Anyway, glad to hear you say you notice.
V8 is an amazing model. 12 steps and i'm getting highly detailed humans, no LORA needed. This is amazing. No other word works.
Thanks! :)
Nice model! How to make a person look full-length?
Generally with 'full body', and add in something about bare feet/shoes/sandals/etc to force it to show more of the lower body.
you can try to add "standing" in positive prompt
"Full body height portrait with feet"
AR = 2:3 or portrait helps too
I've tested so many models, and I get the most consistent and pleasing results from this one. Thank you!
Thank you, glad it works so well for you!
Hi, could someone recommend me an upscaler and its parameters (cfg, steps, etc) I would be very grateful!!
i mostly use 4x_NMKD-Siax_200k
@RalFinger Okay, thank you!!
@ChocoboyMLG What Ralfinger mentions and/or ultramix_balanced. Upscale steps somewhere around half your original steps (slightly up or down depending on how sharp the result is).
I use 4x_foolhardy_Remacri,
Just tried the 4x_NMKD-Siax_200k, and I can't see any or many differences, with a photorealistic image of a cityscape.
Like many, I used Siax, foolhardy_Remacri, 4x RealESRGAN_x4Plus Anime 6B (which I am still testing for everything that is drawing and coloring book style) but after many tests 90% are done with 4x-UltrasharpV10 which suits me for everything ( 6B for the rest). I'm not saying it's the best, I insist, it's the one that suits "me" ;-)
I just posted an image (https://civitai.com/images/19835350) with a workflow whose origin and author I forgot (sorry about that). Take a look at it to see if it suits you.
Thanks a lot for all of your comments!!! I've already resolved it!
@diaserikcg114 Thank you for your kindness and for investing your time to help me ^^
Cool
It's a great model BUT, I tried to recreate an image where the prompt reader said it was made with this model but all I got was crappy blue mist, even deleted all the loras upped the steps upped the cfg, changed schedulers,...but still very strange result, it works great with all my other prompts so I know it is not the models fault, but now I am intrigued what my error could be, I tried everything, best I got was some surrealistic treeroot.
I wish I could add the image or the link without being called on plagiarism. but here is the prompt
(amazing dream:3.5) An amazing of film photo of ((amazing energy ((( plasma))) spectre shaped like a energy tiny creature, on night elven village:1 )) (maximum ultra high definition image quality and rendering:3), maximum image detail, maximum realistic render, (((ultra realist style))), realist side lighting, , 8K high definition, (((8K ultra realist detailed skin))) realist soft lighting, (((realist skin color)))(amazing special effect:3.5)
and here is the negative:
(((3D render))), (high contrast:1.9), (saturation:1)(deformed hand:1.2)(((woman)))(((human face)))(((man)))((((human nun))))
and the build data:
Steps: 40, Sampler: DPM++ 2M Karras, CFG scale: 3, Seed: 1074689079, Size: 832x1216, Clip skip: 2, Created Date: 2024-07-07T17:32:56.0038961Z, Civitai resources: [{"type":"checkpoint","modelVersionId":563988,"modelName":"ZavyChromaXL","modelVersionName":"v8.0"}]
The problem is with the amount of weight you give things. For example, you have (amazing dream:3.5), and many more with such high weights. Generally everything above 1.5, but especially above 2 is a no-go and will result in a bad image. Then you also have a lot of ((((word)))), if you're on A1111 UI (or similar), every ( ) increases weight by 1.2, so if you do that 4 times, the weight is 1.2^4=~2.07.
@Zavy But then why did it work for the original creator, did he use another tool than comfy or a1111 ?
so lose the weights and a lot of the brackets you say, I could try that, thx
@serget2 Could you give me the url to the original image? It probably is originally made on SD1.5, on SDXL these high weights blow up the image normally.
@Zavy Don't have the link anymore, got it from civitiAI though, somewhere in the gallery, I wish there was a reverse image search on CivitAI for this sort of problem. I got the prompts and the data with the SD prompt reader, too bad no username
@Zavy I got prompts and settings with SD prompt reader, so I do not have the link, I am sure it was on CivitAI the other day but I remember the link itself did not give prompts or setting, so I was glad to find it with prompt reader, I think it was a featured image though see if I can find it there
I tested your settings and prompt and got a beautiful evening landscape, where there was a village at the bottom of the frame, and further, behind the hill, a huge strange thing with magical lightning all around. No crappy blue mist or what have you.
@TarXor That is what I should get, Did you change anything on the prompt? Did you post the image?
@serget2 I also get the village scene w/o problem when I paste the pos/neg prompt into Auto111 1.9.3. Usually when I get "blue fog" or "plasma mist" or other (very) weird stuff from pony/sdxl it is caused by the "emphasis mode" setting. Apparently the setting "original" is a deal breaker sometimes. So I added the radio button for this to my quick settings in Auto1111 and set this to "no norm" and everything looks fine. But for your prompt this works either way...
@serget2 I didn't change anything or post anything.
It also happens to me that I get colored spots instead of pictures. This usually happens when I try to torture the neural network a little and mix up unexpected concepts through the A1111 with [tokenA|tokenB] function.
@TarXor yeah but I noticed quite a few times that when I do some stuff in comfyUI my pictures get oversaturated or burnt, never had this blue fog or mist, but I have reason to believe that the weights in ComfyUI are stronger, yet, playing CFG, steps, samplers etc....does not solve it, hence why I am intrigued and willing to fix this, but I haven't made time for that yet (screwed around too long the first time I guess I needed some change of scenery), But I will use your tips and let you guys know if I figure it out
@serget2 Well, there you have it. placing the pos/neg prompt into the super basic workflow of ComfyUI and dialing in the settings (aside from clip skip, don't know where to do it and in comy 2 = -2 ?) I get blue mist. Without mentioning which software you used you had us barking up the wrong tree here ;-)
Drop down those excessive weights from e.g. 3.5 to 1.5 and problem is solved. AFAIK it is (nearly) impossible to get the exact same image between Auto1111<->Comfy<-> onsite anyway. Anyway this is what I get with tamed prompt in Comfy: "Blue mist" investigation | Civitai
@gman_umscht I should have mentioned the too indeed, I am so used to comfy I forget there are others, MY bad, thx for the investigation beautifull pic BTW, I will try it this evening since I am due at my job ATM
Thx for all the hard work guys, you are all top notch
@serget2 BTW pretty sure this is from that set, prompt matches: Image post by greghorfen394 | Civitai
@gman_umscht Another victory for you my friend, it was actually this image https://civitai.com/images/18850751, but yes same creator same prompt, How did you find it? I tried but no luck
It worked you guys lowered the weights significantly and ditched a ton of brackets, now I am seeing better results, will post them shortly, thx for all the help you guys, you're all awesome
I changed your prompt and got this. :D
@serget2 I was dropping images into Auto1111 to remix locally and noticed super high weights of 3.5 on one, and behold it was a creator I already followed - so I scolled thru his gallery and found it.
Because the weights are so high, obviously the emphasis mode will have a large impact, see comparison :
Weight emphasis comparison | Civitai
Not sure how to set the emphasis mode in Comfy, but there will be a node for this I guess.
When I remix it on Civitai in 4 images I get one town from far, one town from ground level, and 2 energy creatures in town, interesting.
@gman_umscht I only changed the weights in the prompt (3.5 to 1.5) and deleted a lot of the brackets to get similar results, but I left my Loras untouched, so it would still be a bit different, And that was all, no special node needed, But my suspicion of Comfy being stronger in weights has now been confirmed, I need to tone things down a lot, otherwise I get oversaturation, A.K.A blue mist
@serget2 There is a node in Comfy where you can use a1111 weights, but I can't remember which one it was.
Ah found it:
Hi I use the civitai image generator and I have no problem with :3.5
I'm not reading all the comments here, cos I'm lazy, but it also could be that the image you are seeing is not the original image, just img2img for correction, hires, inpainting-outpainting, upscaler, detailer, etc. It very well could be that the FIRST original image was made with a different model (or different prompt), and the user used those highly exaggerated prompts to boost the img2img result (again, for upscale, supir / resample, inpainting, hires, etc.), civitai only picks up the last modification's details (because it's a new image, with new metadata). So don't go crazy about this.
Could you pls put recommend settings in [About this version] section?
It's in the description.
The about section is usually a changelog.
As Wizz points out, it's in the description (press 'show more'). The about this version should only be seen as a changelog.
@Zavy clip skip 0-2?
@diamondofdevil835 For SDXL it's always 2 as default. Shouldn't really change that and in most UIs that doesn't get reflected at all because they ignore any other setting.
@Zavy aight okay thanks , and tysm for this incredible Model it can do interesting stuff.
@diamondofdevil835 Do like me: be rebellious and test until it gives out hehe
Ps: 3, 4 and 5 work, not testing more at this time.
Details
Files
zavychromaxl_v80.safetensors
Mirrors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
zavychromaxl_v80.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.