Pixel Art XL
Consider supporting further research on Ko-Fi or Twitter
If you have a request, you can do it via Ko-Fi
Checkout my other models at RetroDiffusion
Best used in conjunction with: Astropulse PixelDetector
Downscale 8 times to get pixel perfect images (use Nearest Neighbors)
Use a fixed VAE to avoid artifacts (0.9 or fp16 fix)
Best results without using, pixel art in the prompt
Tips:
Don't use refiner
Works great with only 1 text encoder
No style prompt required
No trigger keyword require
Works great with isometric and non-isometric
Works with 0.9 and 1.0
Changelog
v1: Initial release
Description
FAQ
Comments (74)
Try this ComfyUI workflow, It works good at all. Use the new checkpoint with baked 0.9 vae. https://ibb.co/MhDPTbj
Anyone got a workflow with the downscaling implemented?
Hi! I didn't understand how to run it. I get regular pictures, not pixelated ones. Can you help me, am I doing something wrong?
This is for SDXL not for SD 1.5
Yup, this is for SDXL.
You can use the example workflow I provided in the examples with ComfyUI
Checkpoint required:
sd_xl_base_1.0.safetensors
Prompt required: parenthesis are optional for "pixel art".
<lora:pixelbuildings128-v1:1> (pixel art)
I use SDXL, no refiner, copy the prompts from the images there, and it is still not pixel art, just regular images.
I'm using InvokeAI 3.
Thank you mate! Amazing work - masterpiece. Try out the img2img - it will blow your mind guys
MattVidProAI did a video on your Lora... https://youtu.be/mPlKKyzvKkE The kid makes awesome videos. Dude, man, whatever. I'm getting to old for this....
I'm 51 and still digging making pixel art, never too old bruv!
This is how I found this lora! xD
This is the lora that finally made me decide to download SDXL. Amazing work!
Might do it for me too. Is it a case of just... Have SDXL's base model in comfy and then grab this? If I swap back to a 1.5 model with 1.5 in the folder, am I still golden?
Here is the json config file for the ComfyUI
Please note you must paste into prompt:
<lora:pixel-art-xl:1> (pixel art)
1.0 lora strength and (pixel art) with extra attention could be a bit to much.
You can try with 0.8 weight, and just "pixel" or even decrease the token strength if the image is to mushy or too burned: "(pixel art:0.8)"
Dude, this is amazing! Now the only thing that's missing is a dithering option and it will be perfect.
You can use the script from https://github.com/devilismyfriend/PXL8
to downscale and dither
How do you use it with comfy?
I love it! On your next update I'd love if you could add 1bit style
I use SDXL, no refiner, copy the prompts from the images there, and it is still not pixel art, just regular images.
I'm using InvokeAI 3.
You need to downscale it like in described in description
@albmagerit785 Turned out InvokeAI didn't support Loras.
Today they released the new version, and everything works now.
Thanks for help anyway.
Any advice on how to use Astropulse PixelDetector with a comfy workflow to avoid having to CMD line each image?
You can use a workflow provided in the astronaut corgi, just drag and drop the image into Comfy and should load the workflow. That has an automatic pixel perfect downscaler
@NeriJS I'm not getting a workflow with that particular image...something I'm missing?
@pxlscript757 Try to click on the "Workflow" and click on the copy button:
https://civitai.com/images/1818851?period=AllTime&periodMode=published&sort=Most+Reactions&view=categories&modelVersionId=130580&modelId=120096&postId=450859
@NeriJS Got it - Thanks!
https://github.com/dimtoneff/ComfyUI-PixelArt-Detector here is the Astropulse PixelDetector code adapted to ComfyUI node
Amazing work! This is the sole reason I am using SDXL.
Any plan to do a similar lora for 1.5? It would be awesome to bypass the much longer time SDXL takes to output images.
Thanks!
I have a couple loras for SD v1.5, sadly this kind results are only possible on SDXL, v1.5 is limited to 64px
you can check those out in my profile
I made one model for pixelart to SD 1.5 if you need, the name is RW pixelart, the difference is that NeriJS's model is pixel art specific while mine does all sorts of sizes(less precise pixelcount)
@leonnn1 Thanks! I'll try it out.
How would I get different variations for my game character? I need a body of a fighter and then use the same pose and body but different outfits, any suggestions?
ControlNet Extension gives you the same pose.
@Fllamber SDXL doesn't work with controlnet
SDXL is good with img2img - close to controlnet alone
If you want a specific pose or scene layout, you can draw a rough sketch in PS or even MSPaint and use that as an input with a high denoise 60-90 (I usually use about 70)
There are also tools that can help, but I've never used them.
When I add this Lore to my ComfyUI it drops error.
How to solve it?
Error occurred when executing LoraLoader: Weights only load failed. Re-running torch.load with weights_only set to False will likely succeed, but it can result in arbitrary code execution.Do it only if you get the file from a trusted source. WeightsUnpickler error: Unsupported operand 83
Hi - I really enjoy using this model, but is it possible to generate images using only a limited number of colours?
Thanks!
Not directly, but you can use an extension or node for color limit :)
https://github.com/dimtoneff/ComfyUI-PixelArt-Detector here is the Astropulse PixelDetector code adapted to ComfyUI node.. there is an option for a reduced color palette
@bamboocha I have tried using this but can't get it setup. I'm fairly new to Comfy UI and am unsure where to put the files download. When dragging the .json file in I get this error:
When loading the graph, the following node types were not found:
PixelArtDetectorSave
PixelArtDetectorToImage
Save_as_webp
Nodes that have failed to load will show as red on the graph.
Any help would be really appreciated.
@readgraves you need to go to ComfyUI custom_nodes, open a cmd and git clone the PixelArt Detector inside. Next time you start Comfy it should have the extensions inside
@bamboocha amazing, will try this in a few hours. I'm going to update to get the new color palette nodes (I bought them for aseprite but will likely use them in Comfy). They look awesome, great work guys!
@readgraves v 1.1 seems to have now grid preview of the palettes... holly shit I am addicted
https://github.com/dimtoneff/ComfyUI-PixelArt-Detector/blob/main/grid.png?raw=true
@bamboocha wow, amazing work. I have been making a load of stuff which takes concept art and produces first passes of character art for dialogue systems (I'm a game developer), this LoRA and colour correction is a massive part of it. It takes me roughly an hour to clean up what I get from Comfy UI, before was taking me around 6 hours! I will grab this update later, thanks for making it, this community is great!
@readgraves then you might want to check this out: https://github.com/dimtoneff/ComfyUI-PixelArt-Detector/issues/2
img2img pixel art base from a 2min drawing to a base. That might also save time and give some ideas to artists.
This Lora Doesn't even show up anywhere in my Lora options in a1111...
Same problem
Me too
Same here
Let me check :O
Is 1111 compatible with XL loras?
Same issue here
update your 1111 and make sure you have an SDXL checkpoint selected.
update: it's showing for me but not working in 1111
@NeriJS Yes it is, I use many of them. Some don't show up in the "browser" but you can still use them.
@chollman82141 New version should be fixed :)
@NeriJS no still doesnt show
@Pentul Make sure to upgrade your install and select a SDXL model before
Make sure you're placing it on the right folder /models/Lora
@NeriJS my bad sorry
Bye bye low res pixel artists lol
Your lack of understanding of art is astounding, lol
hahahahha yep
Dude, this (like any other model) is just another tool in the hands of (pixel) artists. They'll be able to do more and faster than before; they don't need to say goodbye or anything ^^U
Don't be an ass.
@Polygon Yours too lol
(I'm not an artist) Nah, the skill of having direct control of every pixel in your fingertips in real-time is not going anywhere.
Yikes! Imagine taking joy in the prospect of someone losing their livelihood. It's a complete delusion, of course.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.