Like the work I do and want to say thanks? Buy me a coffee or Support me on Patreon for exclusive early access to my models and more!
NightVisionXL started as a lightly trained, photography focused SDXL model that specialized in photographic output. Over the past few months, NightVision has received dozens of trainings and tunings chasing the dragon that is pleasing aesthetic output. With the release of version 8.X, I feel like I'm much closer to my goal of effortless beautiful output that is aesthetically pleasing without requiring confusing negatives or comma separated nonsense.
NightVisionXL speaks natural language. Prompt it like you would describe a scene to a human being, and you'll be surprised by how well it can understand and comprehend your prompts.
I'm trying something new with V8 of NightVision, and I hope the results show in the output. I continue to scrape beautiful imagery that I think will train well, and this particular update is no exception with hundreds of beautiful license free AI images that I've sourced from MJ, Dalle3, Civit, Playground and my own outputs. On top of that, I'm also utilizing HF datasets for license free photography and the LAION-POP dataset, filtered to remove low quality images prior to training. Captioning is done using my own Spicy Burrito captioning tool (available to my Patreon subscribers on discord!) that creates incredibly verbose and effective captions that are essential for good coherent output.
On a final note, I've had multiple requests to "turbo-ize" this model. I have no plans to mix turbo into NightVision, as the licensing on NightVision is nice and open right now with the SDXL licensing, and I'd like to keep it that way. NightVision isn't fast, but I promise you the longer gen times are worth the wait!
This is NightVision XL, a lightly trained base SDXL model that is then further refined with community LORAs to get it to where it is now. NightVision XL has been refined and biased to produce touched-up photorealistic portrait output that is ready-stylized for Social media posting! NightVision XL has nice coherency and is avoiding some of the weird body issues and biases that are starting to plague some of the other photorealistic models. Further, NightVision XL produces rich deep blacks and great evening/night time scenes. It can also produce ridiculously bright output as well!
NightVision XL is capable of both SFW and NSFW output. As with all of my other models, tools and embeddings, NightVision XL is easy to use, preferring simple prompts and letting the model do the heavy lifting for scene building.
NOTE - This version includes a baked VAE, no need to download or use the "suggested" external VAE.
WARNING - DO NOT USE SDXL REFINER WITH NIGHTVISION XL
The SDXL refiner is incompatible and you will have reduced quality output if you try to use the base model refiner with NightVision XL.
Follow me here by clicking the heart ❤️ and liking the model 👍, and you will be notified of any future versions I release. I also need your help with feedback, please please please post your images and your honest feedback below, I will use your feedback and your output to help guide future revisions!
Patreon coming soon!
Description
NightVision XL 0.7.4.3 Release Notes
With this version of NightVision XL, you should notice more natural and normal looking bodies, a bit more facial variety, and much more balanced color output vs 0.7.0.2. Bodies also look better naked, with less noticeable shoulder bone caps on very thin subjects, more natural ribs on less thin subjects, and even more natural features on skin like moles, blemishes and stretch marks, and old people actually look old. Something I've noticed anecdotally but haven't tested in detail is that blood on skin (like a bloody lip or a cut for example) looks much more natural and realistic, and water looks more wet and less like painted on glue, however YMMV.
Bodies should now look much better proportioned and much less waifish and stretched out, especially in tall AR images. To be clear, while it's not 100% fixed, I've put a lot of work into trying to fix the stretchy giraffe necks, weirdly long legs and freaky torsos that all of the community models have inherited from base. While I haven't banished it completely, it seems body balance is MUCH better. Further, while the default output of this model continues to be photorealism biased, the model is much less rigid now, and you should be able to achieve other styles such as line art or anime (and more) with just a bit of style prompting.
TLDR Improvements
body shapes are much more natural, even in non-square ARs
more natural looking skin
less rigid, can produce other styles than photoreal much easier now
less close up portrait bias
Output is less contrasty than 0.7.0.2 and colors are more balanced than 0.6.8.1
09/11/23 Changelog
Backmerges into 0.6.8.1 and 0.6.5.0, and differential merges with (internal only version) 0.7.3.3 which was a WIP model. Added 2 new community loras and an internal fixer lora.
New Community LORAs added
polyhedron_all_sdxl
NaturalBody
New Internal LORA Added
nf_v3-test
Internal LORA Rebalance
Nightshift
FAQ
Comments (27)
NightVision XL 0.7.4.3 Release Notes 09/11/23 (Early Release)
With this version of NightVision XL, you should notice more natural and normal looking bodies, a bit more facial variety, and much more balanced color output vs 0.7.0.2. Bodies also look better naked, with less noticeable shoulder bone caps on very thin subjects, more natural ribs on less thin subjects, and even more natural features on skin like moles, blemishes and stretch marks, and old people actually look old. Something I've noticed anecdotally but haven't tested in detail is that blood on skin (like a bloody lip or a cut for example) looks much more natural and realistic, and water looks more wet and less like painted on glue, however YMMV.
Bodies should now look much better proportioned and much less waifish and stretched out, especially in tall AR images. To be clear, while it's not 100% fixed, I've put a lot of work into trying to fix the stretchy giraffe necks, weirdly long legs and freaky torsos that all of the community models have inherited from base. While I haven't banished it completely, it seems body balance is MUCH better. Further, while the default output of this model continues to be photorealism biased, the model is much less rigid now, and you should be able to achieve other styles such as line art or anime (and more) with just a bit of style prompting.
Note - I am releasing this version for 2 days of civet subscriber early access for early release feedback. If you would like to provide feedback to me directly on this or any of my other models, please head on over to my Discord server SCG-Playground.
TLDR Improvements
- body shapes are much more natural, even in non-square ARs
- more natural looking skin
- less rigid, can produce other styles than photoreal somewhat easier now
- less close up portrait bias
- Output is less contrasty than 0.7.0.2 and colors are more balanced than 0.6.8.1
09/11/23 Changelog
Backmerges into 0.6.8.1 and 0.6.5.0, and differential merges with (internal only version) 0.7.3.3 which was a WIP model. Added 2 new community loras and an internal fixer lora.
New Community LORAs added
- polyhedron_all_sdxl
- NaturalBody
New Internal LORA Added
- nf_v3-test
Internal LORA Rebalance
- Nightshift
I'm sure this model does deliver improvements, but I had some issues with the last version, and ended up rolling it back. The early versions of this model were amazing, but especially the last version just kinda blended into the crowd. I know that's vague, but I don't want to get into a long exposition. Let's just say, early on, I could pick out the Nightvision gens among all the others, but it hasn't been so in the latest version. I didn't want to make a primary comment to say this, just wanted you to know, for whatever it's worth.
@MysteryWrecked Give this one a try, it's a solid improvement over 0.7.0.2, especially in color.
Hi first of all, thank you for the checkpoint. Its one of the best and the one I have chosen to train multiple of my models on. Just a bit of feedback, when generating regularization images from your checkpoint, I have noticed that sometimes/often when i describe the eyes within the prompt, the eyes are glowing like crazy, i don't know how to describe really, but maybe something to look out for if creating 100% realistic checkpoint :)
Hey there - sadly the crazy glowing eyes came up from base and really only shows up if you mention eyecolor (blue eyes, green eyes, etc.) - I've had some luck with lowering the weight when prompting eye color (blue eyes:0.7) but the glowing eyes is something I'm definitely aware of and was just in fact discussing this with somebody this morning.
My suggestion is to never mention eye color in the initial prompt. Instead use Adetailer and mention the eye color there. I use whole person and face and mention the eye color in both. Like "detailed grey blue eyes" or "detailed hazel green eyes". Works perfectly.
Hi another feedback from my testing. When generating prompts with smile. There is very often the same pattern of muscles around the mouth. It looks very often not natural and stretched too far. Some young women look older because of it.
Model rarely deals properly with prompts like joyous laughter, or just laughing in general.
The teeth are decent compared to other models, but still a bit messed up often, especially with smile expressions generation :)
Unfortunately do not yet have access to the new model, so all of this is relevant for the 0.7.2 version.
Hello! I have 2 questions if you got a minute -
1 is what resolutions you like to generate within, as in square, landscape, and portrait
and 2 is up-scaling or hires fix guidelines you would typically recommend.
I love your work, thanks!
also, if you were to NOT use a DPM sampler, what would you go with?
@mumbles Hey there, lots of great questions. For resolutions, I use the "standard" resolutions that are recommended by SAI. This thread over on reddit has them all listed out: https://www.reddit.com/r/StableDiffusion/comments/15c3rf6/sdxl_resolution_cheat_sheet/
For upscaling, I like to use 8x_NKMD_Superscale as my go-to for like 90% of what I do, and 4X_Foolhardy_Remacri for anything else. I know there are better versions for anime and different styles, but honestly, most of the time I'm just generating x/y tests so these are good enough for me.
As for your last question you added on, I'm partial to UniPC, though I do like the output of Restart as well for the non DPM samplers. Euler and its variants belong in the garbage bin at this point IMO, its output is just subpar and its speed is nothing to write home about anymore vs the likes of Restart... Now with all that said, my go to sampler as of late has been DPM++ 3M SDE Karras, it feels like it has the quality of DPM++ SDE Karras and the speed of the 2M series, with 30 steps initial (50% refiner if you're using it, though I ONLY recommend it with NightVision, not with StyleVision or DynaVision as it messes up their styles and makes their output look rather generic), then another 50 steps of HRF at 0.4ish denoiser strength.
I'm happy to give more tips and tricks if you wanna head over to my Discord, SCG-Playground. We've got a small community of long time SD creators who love to share their knowledge. Join us! 👍🏼
@socalguitarist amazing, thank you. I'm omw to discord see ya around
Thanks for your work! I just tested version 0.7.4.3, but for me it's still inferior to ProtoVision. I'm getting weird shaped faces a good part of the time (especially for faces of famous people). Also, all the people seem to have the same elongated, rectangle-shaped lower faces (the rectangle from cheeckbones to jaw sides is too prominent, especially for men). I think it's inherited from the base, it's also noticeable in many other SDXL models.
Meanwhile, I'm getting almost-perfect output from Protovision, so I'll stick to it for now, hope you're still working on it!
Hello. NightvisionXL does not work in my case. I am choosing it from the cozy nest SD checkpoint window and it keeps loading, but then it gets back to another model. Like it can't load itself. Does anyone have the same issue and solve it?
for me when a checkpoint fails to load in my A111 its because the download has managed to corrupt itself , have you tried to redownload it again? might work
@GG_System Yes, I tried that :( After downloading the file only thing to do should be to put it in the models > stable diffusion or I have missed any steps?
@mgjagla452 not sure why its not loading then , sorry
@mgjagla452 Same issue. It appears like it's about to load, then just loads a different model (the last one I'd used).
guys, returning to old Nvidia drivers seemed to work for me.
me too.
hi, i have this error :
modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
I've followed the recommendations, but nothing changes.
you need to add --no-half-vae to your launch options, should clear that right up. If you still have trouble, pop on over to my discord server, happy to help out :) Link is in the comments here on the model.
I use this as a refiner, honestly it makes wonders !! i use it at 0.6 refine (150 steps), even worst images become amazing
wusshh ? excellent ! .. then so which SDXL you have found best to use with this pseudo Refiner ?
and this sounds like refiner can be changed by any SDxl model also ?
do you experiment that way ?
But what ? 150 steps ? every image could be ok after so many steps ! ?;) why so much ?
150 steps... bit excessive? 50 is usually extreme overkill.
@TheP3NGU1N its just a preference in how the final image looks. specially when using euler a, the image keeps changing the more the steps are , @akozoom usually depends on what i wana see at the end, but mostly cardos, Niji 5.1 and brightprotonikino
@7sanal7san thank you ! and in so long steps, you may try DDim which goes forward a final same image as euler_a final in less time (55steps is good). But well i suppose you've tried it also ;) and for sure euler_a but Kdpm2_a is funnier for searching less smoothy images than euler_a
Does this model have mammoplasty like artifacts on medium breasts like protovision has?
Details
Files
nightvisionxl_v0743ReleaseBakedvae.safetensors
Mirrors
nightvisionXLPhotorealisticPortrait_v0743ReleaseBakedvae.safetensors
nightvisionxl_v0743ReleaseBakedvae.safetensors
nightvisionXLPhotorealisticPortrait_v0743ReleaseBakedvae.safetensors
nightvisionXLPhotorealisticPortrait_v0743ReleaseBakedvae.safetensors
nightvisionXLPhotorealisticPortrait_v0743ReleaseBakedvae.safetensors
nightvisionXLPhotorealisticPortrait_v0743ReleaseBakedvae.safetensors
nightvisionXLPhotorealisticPortrait_v0743ReleaseBakedva.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.