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!
Meet ProtoVision XL! This model is my "Bob Ross" model, a happy accident that happened when I goofed up some merge settings on my DynaVision XL model and stumbled upon some magic.
So what is ProtoVisionXL? At its core it's a custom built version of NightVision XL, which I then pile multiple fabulous LORAs on top of to create a unique look.
Like all of my other models, merges, embeds and tools, ProtoVisionXL is designed to be easy to use and excels with simple prompts, but is very coherent and can build fabulous scenes with exquisite detail.
ProtoVisionXL is a portrait model, it favors showing human subjects. That's not to say you can't get other art styles, creatures, landscapes and objects out of it, as it's still SDXL at its core and is very capable.
WARNING - DO NOT USE SDXL REFINER WITH PROTOVISION XL
The SDXL refiner is incompatible and you will have reduced quality output if you try to use the base model refiner with ProtoVision XL.
Follow me here by clicking the heart 鉂わ笍 and liking the model 馃憤, 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
LORAs used for the initial Alpha version of ProtoVisionXL
add-detail-x
AdvancedEnhancerXL
DetailedEyes_xl_V2
MovieStillXLV7
Neonify
offset_example-lora_1.0
photorealistic_slider_v1-0
Samaritan 3d Cartoon SDXL
dwzstyle-000018
wowifierXL
Description
ProtoVision XL Release Notes 09/29/23
For this version of ProtoVision, I did a ground up rebuild to get the latest versions of DynaVision and NightVision worked into ProtoVision to bring in the latest fixes from both models. What this has accomplished is a more flexible version of ProtoVision that is somehow greater than the sum of its parts. My model testers have been going gaga for this version, and if I would say the BIGGEST improvement is background coherence. You're going to find backgrounds are incredibly detailed, and "make sense". Further, this version is capable of incredible ultrawide landscape output, and the versatility is through the roof.
I hate saying this when I discuss my models but, you gotta try it yourself - it's goooooooooooooooood 馃槈 - easily my best model release yet!
ChangeLog Version 0.6.2.0
This was a simple rebuild. Only components are PVXL 0.5.2.0, NVXL 0.7.4.3 and DVXL 0.5.3.4 - No LORA updates on this one, just pure merge magic.
FAQ
Comments (33)
Without a doubt, this is my favorite model in the world. I was a user of playground AI, where they have this model available to test. I fell in love with the results so much that I decided to buy a graphics card so I could do it locally. The thing is this: when I try to replicate the artwork with the same parameters and seed, I can't replicate the results. They don't even look alike. Could it be a different specific version? or another vae? What could I be doing wrong? thank you for your hard work.
Hey, head on over to my discord server (link is in the model comments) and we can troubleshoot. Is Playground AI charging access to my model? this is the first I've heard they're using it over there.
@socalguitarist聽no, it's free. But now I have a gpu I want to try it on my pc. To use loras, controlnet, after detailer, etc.
One thing I've noticed is that I get different results if this is the first model loaded vs. if I loaded after another model. The problem could be VAE related, but I get good results either way, it just confused me when I couldn't reproduce an image I had made the previous day (until I did the other model load/reload proto trick).
YES! An upgrade to my favorite model! Thank you so much for this! Any update log for the changes?
PROVISION LET'S GOOOOOOOO!!!!
ahem.. PROTO VISION!!!!!!! YOOOOOOOOOO!!!!
w
You're making some killer models. I have a special request. How about a model based on high resolution images of art from wikicommons. Things like Signac heavily impastoed pointillistic brushstrokes, Goya's Capricho etchings with lush aquatint and drypoint scratches, Rembrandt's balance of transparent glazes and ropey brushstrokes, Rothko's Color-field scumblin etc. No one has done a decent art checkpoint with XL and there's only a few hit and miss LORAs here and there, and those LORA's rarely work well with most checkpoints. Just a suggestion. I bet it would be more popular than people expect.
My suggestion is that if you have very specific things you wanna do just train your own LoRA! :D There are millions of tuts out there now and many articles here on Civit that can get you started!
@olivetty聽Actually, there are lots of great LORAs already, almost to the point where I don't need to train my own. HOWEVER, there are not that many great XL checkpoints to run them with that actually get decent artistic results especially in the 2d and monochromatic world. Especially checkpoints that have high CFG resistance for complex prompts or that stay cohesive with high step counts. There is a plethora of these with 1.5, and XL rarely exceed it for my needs. I just want someone with the hardware capable to prove me wrong that 1.5 is better than XL (other than for cartoony, saturated, photoreal, Pixar stuff).
@Crowyote聽I don't want to promote my own stuff on someoneelses stuff but please try my Realities Edge Anime. Don't be fooled by the ANIME in the name, it is very capable at doing art in any form. Just click my username :) New version for that coming soon as well!
@Crowyote聽Also, XL is still in its infancy and we've not gotten a NovelAI leak for XL yet, that checkpoint alone catapulted the art-centric ckpts during the beginning of 1.5, if you recall? Good luck and patience!
@olivetty聽Actually, I've used Realities Edge quite a bit. It's excellent, and I've gotten some good results. Between CartoonXL and your's I'm getting the best/most interesting results in the 2d space. My biggest issue is I haven't found a GUI that incorporates Img2Img with XL and LORA's to my liking. For now, I'm waiting on NMKD to improve his GUI as I find it the most user-friendly. I'm avidly awaiting your next checkpoint!
@Crowyote聽May I ask, what is wrong with img2img in A1111 when using XL and LORAs? I use it daily with and without controlnet and works fine!
@olivetty聽To be quite candid, I just don't like the layout of Auto1111's GUI vs NMKD's GUI.
Particulary, I like how NMKD has a list/menu for LORAs that you can select and check the ones that you want, and not have to pollute your prompt with activations like <lora:etc:etc>, and instead just use trigger words OR not (the LORAs still work albeit much more subtly). I tend to stack up as many as 10 LORAs with very diluted strength to get the look I'm going for.
The only thing I do like about Auto are the preview of progress of image generation, and the ability to try newer samplers; but, with the same settings as I use in NMKD I get weaker/overall less transformative results to an Img2Img in Auto. NMKD is about to release his new 1.13 beta which hopefully will work perfectly with SDXL, Img2Img, and all LORAs therein. I still like his 1.12 for SDXL checkpoints and Img2Img, it's easy fun and fast. I tend to stick to 1.11 for SD1.5 and LORAs with Img2Img, so easy and such profound results.
A new protovision!? Oh goody!
ProtoVision XL Release Notes 09/29/23
For this version of ProtoVision, I did a ground up rebuild to get the latest versions of DynaVision and NightVision worked into ProtoVision to bring in the latest fixes from both models. What this has accomplished is a more flexible version of ProtoVision that is somehow greater than the sum of its parts. My model testers have been going gaga for this version, and if I would say the BIGGEST improvement is background coherence. You're going to find backgrounds are incredibly detailed, and "make sense". Further, this version is capable of incredible ultrawide landscape output, and the versatility is through the roof.
I hate saying this when I discuss my models but, you gotta try it yourself - it's goooooooooooooooood 馃槈 - easily my best model release yet!
ChangeLog Version 0.6.2.0
This was a simple rebuild. Only components are PVXL 0.5.2.0, NVXL 0.7.4.3 and DVXL 0.5.3.4 - No LORA updates on this one, just pure merge magic.
Join us over at SCG-Playground, my community discord server where we share our art to our heart's content!
Sorry for putting these up late!
Thanks for the work of socalguitarist. This is the best model of texture I have encountered since I tested more than a dozen SDXL. Although the characters generated by the latest v0.6.2 are more realistic, I feel that the color of v0.5.2 seems to be better.
I've been waiting for a long time to receive this update, and it's fantastic! I can't use any other checkpoint because it's not as good. Thank you!
Can you guys tell me why with this model when I activate a lora in the prompt it gives me Networks with errors?
Make sure the LORA you are using is an SDXL LORA and not a 1.5 (or 2.X) version. If you're still running into trouble and you've confirmed it's an SDXL LORA you're trying to run with ProtoVision and it's still not working, come on over to my discord server (link in the model comments) and we'd be happy to help troubleshoot!
This will be the first bad review about this wonderful model. With each new version, the model has less animation and more hyperrealism/photorealism. In 0.6.2 I don鈥檛 understand how to force it to always generate cartoon characters (any slight change in the prompt can generate photorealism.
same here, all I want is to do 2d. All the models tends to photorealism. No 2d friendly models for sdxl yet.
This is a positive review for me.
Hey there, would you mind pinging me on discord? I'd like to pick your brain on the style you're looking for, as I don't like to hear it's not working how you want. I'm thinking I could potentially create a divergent version that focuses more on the animation aspect, and I'd love to see any examples you can provide (as it will help me get the feel right for the training) - anyway, I hope you reach out!
By far the best model I have used so far, already look forward to your future work!
This is hands down my favorite model!
I've been trying it out on Rendernet[.]ai and it can create virtually anything...
Hats off to the creator!
Hey man!
I am working for a company in the SD space and we love your work:) Is there any way to contact you about a collaboration/partnership?
You can also send me a message on either:
Twitter @bozicrok
Discord @rok500
Thank you & keep creating<3
very nice. thank you from hong kong user
I'm not sure why you recommend not using the SDXL refiner with this model. I've tested it, and it clearly adds detail that you don't get without it, and works fine.
Excellent, Gj => good job, Great Job...
nothing to add :)