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!
DynaVision XL was born from a merge of my NightVision XL model and several fantastic LORAs including Sameritan's wonderful 3D Cartoon LORA and the Wowifier LORA, to create a model that produces stylized 3D model output similar to computer graphics animation like Pixar, Dreamworks, Disney Studios, Nickelodeon, etc.
DynaVision XL is capable of both SFW and NSFW output. As with all of my other models, tools and embeddings, DynaVision XL is easy to use, preferring simple prompts and letting the model do the heavy lifting for scene building.
WARNING - DO NOT USE SDXL REFINER WITH DYNAVISION XL
The SDXL refiner is incompatible and you will have reduced quality output if you try to use the base model refiner with DynaVision 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!
Description
DynaVision Version 0.5.3.4 Release Notes
This release is specifically to update the version of NightVision XL under the hood, and has been rebuilt with NVXL 0.7.4.3 as its base along with some careful back-merging to retain the DynaVision feel. I'm happy to report that just like the latest version of NightVision, 0.5.3.4 is less rigid than past versions of DynaVision, and while it retains the DynaVision charm (and big eyes), it is more flexible when prompting different styles now, and really puts a unique spin on things like anime and wildlife art. Color should be more neutral overall. This version operates best in the 0.25 to 0.6 range CFG - you can go higher, but things will start to look a little cooked if you go up much past 0.75.
Also, while it's not perfect (and likely never will be), I've decided to remove the beta tag from this version, and will from this point forward be maintaining a more relaxed release schedule, specifically to give the internal builds plenty of time in the hands of my testing crew. If you'd like to see some of the work that goes on behind the scenes for making models, please join me over on my discord server, #SCG-Playground where we discuss generation techniques, share prompts and wildcards, discuss the latest events in the AI space, and post to our hearts content in the Flood channels, where there is no limit to the number of images you can share. Join us! 💖💖
Changelog
Fresh rebuild using NightVision 0.7.4.3 as a base
backmerged both 0.4.1.1 and 0.3.7.1
rebalanced SDXL 3d render style xl and Samaritan 3d Cartoon SDXL LoRAs
Known Issues
Hands
Male Genitalia
shirt nipples (this should be better now, but still happens on occasion)
logos and signatures - again, should be better, but still happens on occasion, especially with styles that typically have signatures, such as oil painting styles
Crooked eyes - This is something picked up from 0.7.4.3, it's rare, but when it happens, you know it, it will just render each side of the face differently and you'll end up with goofy crooked eyes. I'm trying to suss out the cause, in the meantime, just reroll if you get goofy eyes.
FAQ
Comments (30)
DynaVision Version 0.5.3.4 Release Notes
This release is specifically to update the version of NightVision XL under the hood, and has been rebuilt with NVXL 0.7.4.3 as its base along with some careful back-merging to retain the DynaVision feel. I'm happy to report that just like the latest version of NightVision, 0.5.3.4 is less rigid than past versions of DynaVision, and while it retains the DynaVision charm (and big eyes), it is more flexible when prompting different styles now, and really puts a unique spin on things like anime and wildlife art. Color should be more neutral overall. This version operates best in the 0.25 to 0.6 range CFG - you can go higher, but things will start to look a little cooked if you go up much past 0.75.
Also, while it's not perfect (and likely never will be), I've decided to remove the beta tag from this version, and will from this point forward be maintaining a more relaxed release schedule, specifically to give the internal builds plenty of time in the hands of my testing crew. If you'd like to see some of the work that goes on behind the scenes for making models, please join me over on my discord server, #SCG-Playground where we discuss generation techniques, share prompts and wildcards, discuss the latest events in the AI space, and post to our hearts content in the Flood channels, where there is no limit to the number of images you can share. Join us! 💖💖
Changelog
Fresh rebuild using NightVision 0.7.4.3 as a base
backmerged both 0.4.1.1 and 0.3.7.1
rebalanced SDXL 3d render style xl and Samaritan 3d Cartoon SDXL LoRAs
Known Issues
Hands
Male Genitalia
shirt nipples (this should be better now, but still happens on occasion)
logos and signatures - again, should be better, but still happens on occasion, especially with styles that typically have signatures, such as oil painting styles
Crooked eyes - This is something picked up from 0.7.4.3, it's rare, but when it happens, you know it, it will just render each side of the face differently and you'll end up with goofy crooked eyes. I'm trying to suss out the cause, in the meantime, just reroll if you get goofy eyes.
@socalguitarist I think this one has lost some of its charm - check out this comparison I made:
Image posted by zer0TF (civitai.com)
I much prefer 0.4 to 0.5 - I'm hoping 0.6 brings back some of the unique flair and style that 0.4 had!
Excellent SDXL model. I featured this model on the artist style browser site. You can compare it's output to vanilla SDXL for 1,000+ images.
thats awesome lib,
is it or trial with limitations?
No, there's no trial, no limitations, and no accounts. What you see is the same as when everyone sees. In fact, you can download all the data, code, and images from Hugging face if you like.
The best SDXL model for me so far! Huge thanks to the author for a job well done꒰ღ˘‿˘ற꒱❤⃛
Heads up, you can now create images directly on Civit! Just hit that little create button in the corner. I've added all of my models to the pool, so if you don't have the hardware to use SDXL locally, try it out here on Civit!
I've not tried running anything through CIv before. When I click the play button to run this model I only have 2 online options, one is Diffus which goes to some website graviti and the other is an iphone/mac app (which I don't have). Am I missing something, or is that correct?
This model is too good. I've been using just base SDXL for a while to get to know it. Compared to that, using this model feels like I'm cheating (´。＿。｀)
I'm getting an error loading this model:
Stable diffusion model failed to load *** Error completing request *** Arguments: ..... safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer
what kind of upscaler can we use for this model? I am getting an error for almost all of the popular upscalers. Thanks in advance.
Try to use Ultimate SD Upscale in the Img2img tab
@shortsqueeze thank you
If you're getting upscaler errors using HRF, you likely need to add --no-half-vae to your launch options. Pretty common issue with SDXL.
4x_NMKD-Siax_200k, 8x_NMKD-Superscale_150000_G, lollypop, 4x_foolhardy_Remacri & R-ESRGAN 4x+ Anime6B are all really good upscalers for this particular model. If you go here you will be able to dnld a large variety of the legacy & new upscaler models - https://civitai.com/models/147838/ultimate-esrgan-upscalers-collection-bundle-aio or go here to pick & choice which ones you'd like to dnld - https://openmodeldb.info/
@socalguitarist is right about the --no-half-vae being added to your webui-user.bat commandline-argu.
Thank you all, this issue got resolved. I realized I have to pick vae, render an image, then pick "none" under vae.
this does not work.
Error Traceback (most recent call last):
File "diffusionbee_backend.py", line 325, in <module>
File "convert_model.py", line 80, in convert_model
ValueError: The model is not supported. Please make sure it is a valid SD 1.4/1.5/2.1 .ckpt/safetensor file
[42370] Failed to execute script 'diffusionbee_backend' due to unhandled exception!
ValueError: The model is not supported. Please make sure it is a valid SD 1.4/1.5/2.1 .ckpt/safetensor file
But it is non of those, its SDXL.
Are you using a UI that doesn't support SDXL?
You need to update a1111 to version 1.5.0 or above to be able to run SDXL .ckpt/safetensor models. (If you add git pull to the beginning of your webui-user.bat file it will automatically update your UI to the latest version every time you run it) I've been using this particular model since the day it was released & it works brilliantly, granted the more VRAM you have the better equipped your machine will be to handle the new larger sized SDXL models.
I'm not sure if Diffusionbee has SDXL yet, but if you're using a mac, drawthings supports it.
I really like this! Thank you for the great work. One comment, however: Nipples seem to come out a bit funky. Anybody has some advice on fixing this?
I have noticed this as well. Sometimes they attack through the clothes or show up as buttons on top.
New version coming in a few days that should hopefully be an improvement here.
The longnips lora plays well with this generally speaking.
does SDXL generally take a long time to render then sd1.5 models ? mine is taking, at 512 it was a mess like overtrianed thing, at 1024 its not ready to give out anything
I heard SDXL works way slower on automatic1111 than on ComfyUI. Maybe you are using the former and that's why it's slow.
Depends on your hardware but yes XL is slower and uses much more vram. To give you an idea 512 on SD1.5 is 6-7 it/s, 1024 on XL is 1-1.5 it/s on a 4060 Ti/Automatic1111.
I do no know Comfy at all.
Sounds like a problem with VRAM. When VRAM is full you either get an error and no output OR it just takes a really long time, it all depends on your driver. ComfyUI works better with 8 GB VRAM (which is the absolute minimum), so try that first.
Performance will increase few times and more
Try to use in bat file webui-user.bat:
set COMMANDLINE_ARGS=--medvram-sdxl --no-half-vae --xformers --update-check --update-all-extensions --autolaunch
Just a heads up, I have a new version of DynaVision in internal testing. Everything is looking great so far, and I expect to have the new update out no later than Wednesday, barring my testing team finding any major issues.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.