Create STUNNING 🏆 AI Images with SDXL 1.0 & ComfyUI: Step-by-Step Guide

[@JT-Works]

Glad to see you are posting again. I figured I would throw a comment in to help with the algorithm.

[@RealityCheckVR]

Hey buddy! I always appreciate your support brother. We should definitely connect and catch up soon. I bet the little one would like the mini trampoline I've got set up for Danil. 🙌

[@zenmasterjay1]

Dude I'm not even a minute into this thing and I'm lmao😂😂😂 u prob made my day thnx😂🎉🎉🎉

[@JolieKim2795]

This vid is highly recommended for beginners, although A111 is still very good, the creative art style has drawn me to explore comfuyi.This is the easiest-to-understand video for beginners, with clear images, slow and detailed instructions. If I have to sit and watch how to pull each wire (the supposed set-up of comfuyi post-production), it will probably be a long time before I can use comfuyi. thanks for sharing

[@RealityCheckVR]

Glad it could be helpful! I've got some more advanced ones coming, still, the comfyui customizations are super easy once you get started. I'd say it took me about 5 minutes of learning before I started thinking of all sorts of new possibilities. I'll have an update showing this exactly here in the next week! 👍

[@XalenMaru]

First of all can't say how great it is to see you making videos again! Also wanted to say thanks for this tutorial. I tried to get ComfyUI running before since I'm using the DreamshaperXL10 model but couldn't get it working. The instructions you provided fixed my issues.

[@RealityCheckVR]

💖🙌 I'm excited to have some time again! I just downloaded the DreamshaperXL model last night myself. Can't wait to begin training my own models with SDXL. I always appreciate the feedback!

[@AnitaPotterProductions]

I just got all of this today before I found your video. Nice to see someone else with the same GPU as I have. Just with the default prompt it took around 10 minutes to generate an image. No sure how long this prompt I'd once used in Midjourney will take to complete (wish there was a timer or something) but here's hoping I get a better image than what MJ gave me LOL!

[@BerserkerX805]

Thank you for making this video. I plan to try it out.

[@micbab-vg2mu]

Great video, thank you. Stable Diffusion XL 1.0 has potential. I think that after adding LORAs and ControlNet and other options from SD 1.5 - it will be a game changer.

[@RealityCheckVR]

I'm SO EXCITED! 😄

[@blenderbuffs]

Thank you! great tutorial!

[@RealityCheckVR]

Glad it was helpful!

[@Dizasteroner]

Thanks u saved my day ^^

[@ilokikoval]

Thanks, great content. Would be cool to have a tutorial on using ControlNets like OpenPose with ComfyUI, as well as creating bigger resolution images with outpainting.

[@RealityCheckVR]

Good idea! I've got some Gaussian Splatting videos i'm working on right now, still, there is so much more with image generation that I want to get into!

[@paulhorn24]

Any experiences using SDXL with Apples CPU/GPU/RAM System on a Chip (SoC) compared to Intel-x86 based hardware (or AMD based hardware)? Thanks!

[@RealityCheckVR]

I know a few people that started to do this, unfortunately the speeds were pretty slow. I have one buddy who just uses online services for a couple dollars a day to host his GPU and run his software.

[@paulhorn24]

@@RealityCheckVR Okay, thanks. Apple‘s Mac studio has 76-core GPU with 192 GB memory and 800 GB/s memory bandwidth. It includes a Media engine with two video decode engines, up to four video encode engines, and up to four ProRes encode and decode engines. The CPU has 24 cores. This configuration probably won’t have any problems or speed issues, but one needs to order it and of course has to pay it in the end :-))))

[@RealityCheckVR]

Oh yea! The last place I worked bought one of those but I never had the chance to play with it. SOO much potential in that tiny little white box 😅

[@joeyhandles]

Is there an improved workflow out?

[@sidheart8905]

Comfy ui is best for sdxl its so fast and amazing and does not overloads gpu like others

[@RealityCheckVR]

agreed!

[@ginogiori5285]

Great video and great workflow. However, I don't know how to clear pictures/cache in the IMG2IMG, so it keeps defaulting to the picture and never the written description!

[@RealityCheckVR]

I'm not sure if I understand the problem exactly, do you have screenshots available?

[@ginogiori5285]

I figured it out. Please send us any revised workflows. It would be impossible for us newbies to use this software with the basic workflow offered. Your workflow is 1000 times better. Your help is INVALUABLE. I do not understand how to make a rectangular picture. Even if I change sizes in txt2image, and size and steps, it always spits out a large square picture. How would I do 24"x36"?

[@ImAlecPonce]

Thanks!!! I was wondering where to get the json for sdxl!!! (And understand how they work a bit better)

[@RealityCheckVR]

Glad it could help! I'll work on making updates as I continue to learn myself. 👨🏼‍💻🙌

Can you do a test is SDXL understand "award winning", I tested it in SD 1.5 and it was like that model did not know what "award" was and there was no larger diference between the images. I use A1111 so it is harder to do a test where two images with sloghtly diferent prompts can be rendered at the same time.

[@RealityCheckVR]

I'm on it!

[@jonorgames6596]

Thanks for the video! Can one use AMD gpu?

[@RealityCheckVR]

Definitely! Check out the AMD section on the GitHub - github.com/comfyanonymous/ComfyUI#amd-gpus-linux-only

[@jonorgames6596]

@@RealityCheckVR Sadly only about 512x512 works without running out of vram :( with 12gb. Though, cpu rendition works but it takes about 1 hour per image!

[@ZeroCool22]

How i can add a Lora in this workflow?

[@RealityCheckVR]

Try adding into your prompt to test it. Hope that helps! 🙌

[@DanielPartzsch]

Thanks. Can you please share a link to the face swap tool you've mentioned at the end of the video?

[@RealityCheckVR]

This is the tools that I use for photos - insightface.ai/, I can make sure to touch on this more in future videos too!

[@DanielPartzsch]

@@RealityCheckVR Thank you

[@michealubioyvut]

Hi dude your videos all very useful for me , i have one doubt about SDXL 1.0 AI images i can use commercial purpose

[@RealityCheckVR]

Yes, you can, they said in their launch that is usable for commercial use! 💯

[@michealubioyvut]

@@RealityCheckVR thank you dude I am not use SDXL 1.0 in directly but I can clipdrop website pro version.

[@ViniCavin]

Great vid man! Which tools did you use to make the face swapping and voice changing? Really cool!

[@RealityCheckVR]

I used ROOP for the FaceSwap and RVC for the voice cloning, I'll make a quick tutorial in the next week to show how it's done!

[@ViniCavin]

@@RealityCheckVR That would be awesome! This is really cool to create content :)

[@arthurmartin9398]

Thx for sharing this. Does this work in other languages too? Like in German?

[@RealityCheckVR]

I know that it works in French, German and Spanish for a large amount of prompts, even so, it's based on the training data and using the same prompt in different languages will surely result in different prompts. Might be something I explore in a further video! Hope this helps!

[@arthurmartin9398]

@@RealityCheckVR thx, i will try it, while watching your video again. Have a great day

[@asitmishra8431]

Hey Can You Teach How you did that voice of Thor?

[@RealityCheckVR]

Absolutely! I'll work on getting a simple tutorial made ASAP! :D

[@abdullahmansoor1]

Great tutorial! Thank you so much! How can I add lora section to it?

[@RealityCheckVR]

Try adding into your prompt to test it. Hope that helps! 🙌

[@grcunning]

Thanks for the content. I do wish your mouse cursor was bigger, it is hard to see.

[@RealityCheckVR]

I accidentally had it hidden on the main screen in OBS, ill have that fixed for next tutorial! Did you get stuck anywhere that I can help?

[@MarcoEscobedo10]

I like your workflow. Can you add Lora to the .json file please

[@RealityCheckVR]

new video just dropped with a better workflow! kzbin.info/www/bejne/iIGWc5Weqpapjqc

[@globalmodelsdotbiz8965]

1 - You can use 0.9 refiner with base 1.0 and may produce other, preferred, results, 2 - The word is that at occasion 1.0 base plain produces a better results than with the refiner .... expiriment!

[@RealityCheckVR]

Experimenting!!!

[@ismailtibba]

Split the base model from the refiner and you will get much faster results

[@calebcriste]

Tell me more....? 🤓

[@ismailtibba]

@@calebcriste render the image without refiner like all we use to do, and use the refiner in img2img setup this worflow save me almost 40% in render time.

[@RealityCheckVR]

I'm not at my PC right now, I assume this is as simple as configuring the connections through ComfyUI?

[@thatwittyname2578]

@@RealityCheckVRI assume he is talking about encoding the finished image back into a latent then running that through the refiner

[@ismailtibba]

@@RealityCheckVR Yes for sdxl base model workflow is simle as txt2img for refiner use workflow fo img2img

[@joeduffy52]

Looking at your prompt, you asked for *an* elvis which is probably why you got the car. As we know Elvis is/was unique!

[@RealityCheckVR]

you're absolutely right, you can see if you drop the images in that it's the only difference, lol

[@Pan-Nikt]

Hi where I can download more models for Comfy UI?

[@RealityCheckVR]

One of the best models is DreamShaperXL which is available here - civitai.com/models/112902?modelVersionId=126688 - surely there will be many more models added in the coming weeks. Several other websites host them as well, still, this is my most frequented one.

[@tristanmoreau1956]

you know that you don't have to install comfuy, since the version 1.5.1,you just have to put your model base and refiner in you model stable diffusion and run it txtoimg then imgtoimg for refiner

[@RealityCheckVR]

right! I plan to make a video showing this method as well. I have a lot of friends that do not have A1111 or other GUI's installed so I believe ComfyUI is the easiest to get going from scratch, thus I made this little vid. I'd like to update my training models and make some videos about that as well if you find any of that interesting!

[@SuperSam_GER]

Just don't work as easy as you doing in the video. Just after downloading and starting it says: Load Image: - Custom validation failed for node: image - Invalid image file: 00005-1840926806 (1).png UpscaleModelLoader: -Value not in list: model_name: 'None' not in [] LoraLoader: -Value not in list: lora_name: 'sd_xl_offset_example-lora_1.0.safetensors not in [] And you told loras are not needed

[@p4duedits]

i did everything right but the refined version was mising

[@RealityCheckVR]

can you elaborate a bit? I'd love to help if I can! Did you drop the .json file into ComfyUI>?

[@p4duedits]

@@RealityCheckVR yes I did, i tried once and I just got the base version but before that I got warning with some red text appeared and i clicked ok

[@p4duedits]

new problem it disconnects right before the image is generated

[@elrika5302]

Why is it so dammn slow to extract 😩

[@RealityCheckVR]

Get it working?

[@1Garrett2010]

Did someone test SDXL 1.0 with less than 16gb VRAM? It works?

[@RealityCheckVR]

I'm using a 1080ti that has about 11gb built-in. CPU only works too but takes forever

[@1Garrett2010]

@@RealityCheckVR Really interesting, I have a 980ti with 6gb and I'm thinking for an RTX 4060 ti 16gb when available or a good "old" RTX 3060 12gb

[@chrisjohnston8952]

Does DreamBooth not work with SDXL yet?

[@RealityCheckVR]

I haven't tested it yet myself, the difference is now you have a base model as well as a refined model that must be trained and exported. I have seen a few videos coming out about it but haven't tested it yet. Let me know what you find!

[@CMDRRustyDog]

"32k definition" no chance of getting that in a 1080 image. Not enough pixels, so it's redundant. Same with 4k, 8k etc. People use so many useless prompts in their images. I had one with '8k, 4k' :) I can just hear the AI renderer saying "make up your mind!"

[@RealityCheckVR]

no doubt, I copied it from an image I found on civitai.com/ and didn't even use the other 15 random words it had attached to it.

[@CMDRRustyDog]

@@RealityCheckVR I'm the same :) I look at prompts from other images and I have to trim it down to the stuff that might make more sense. I've seen some bizarre prompts I can tell ya.