#### Links from my Video #### Lighting Based ControlNet civitai.com/models/80536?modelVersionId=85428 Negative Embeddings: docs.google.com/document/d/1ODIUsxT0Am1QX0BDvtUbXYIw7LHFS-TzVxjyAShhQBw/edit?usp=sharing Install ControlNET: ua-cam.com/video/zrGLEgGFJY4/v-deo.html #### Join and Support me #### Buy me a Coffee: www.buymeacoffee.com/oliviotutorials Join my Facebook Group: facebook.com/groups/theairevolution Joint my Discord Group: discord.gg/XKAk7GUzAW
Your content is insane, and the community around AI (both LLMs and image generation ones) is one of the best. We get so many new things every single day, that it's already impossible to keep up. You have my thanks for making it easier.
Really cool, I've seen Clipdrop has a dyamic relight tool for editing multiple lights based on depthmaps. It would be cool if a tool like that would be integrated in A1111.
It's quite amazing how whenever I think, "Hmm, I need to know about X in Stable Diffusion/ControlNet/LORAs..." and bingo, Olivio has it covered! I was just thinking how it would be easier comping stuff into AI-generated backgrounds if I could get the background lighting to match my lighting setup here in the studio at home. And this looks just the job for that. Your last example is just really, really clever. Shows once again how old-school comping knowledge is the key to a really good AI image. Bravo!
This is nice but I find that controlling lighting is pretty consistent via prompt, like color+ backlight, spotlight rimlight underlight even flashlight works,. also mentioning daytime or weather conditions like , day night dawn dusk overcast sunny hazy foggy dusty etc.
this is great that it has a model, I've been doing this for months using gradient maps and img to img. Might make a video of that as an alternative method. Great video as usual Olivio thanks.
It's kind of a nice CN addition, but after playing with it a while, i dont see the point using it. It is way more easy to add some light in photoshop where i have full control over what im doing than trying thousand times to get the right light from the AI, which i need to adjust in PS afterwards anyway. I mean you create a light map in PS, why just dont using this map with your final image in PS ;)
Been subscribed for a while now and loving your content. Have learned so much more. If it's even possible to render a car with properly round wheels (I've spend many hours trying), would you be able to do a video showing how it can be done? Keep up the great content.
Yes, a lighting editor like the OpenPose editor would be a dream for people like myself who use SD only for inpainting and comping. You'd just add the lighting like in your home setup/studio and then generate a background that is a really close match to your own lighting, making the relight so much easier. Even this already gives me incredible results.
😃Hey love your videos! Was wondering if you might be able to do a step-by-step guide on how to uninstall and do a clean install of auto1111? Something happened to mine and it just stopped launching. Tried to reinstall it and I still have issues with it not launching. Last thing I did was add ROOP and update it. After that everything went bellyup :(
I've been experimenting with this for awhile now and find the workflow quite finnicky and limited to be honest. You can get some interesting results for sure but if you'd like to "re-light" any of your older images then you're in for a rough time. I think a lot of people commenting here hasn't actually tried this out so be warned, it's not quite as easy as it might seem =)
Yep! It's real bad. It's interesting though, and promising as a concept. But it seems basically to be a 100% lock on all transformer layers that you just scale back and kind of hope it moves your image in the right direction. It certainly doesn't do what it says on the tin.
Unrelated question/ tutorial idea: Is it possible to combine the texture from person A with the features of person B? Examples: a female version of the rock band Kiss or Alice in Wonderland with Alice Cooper's makeup.
Open pose never works for me. It puts the persons limbs in the right spot but always like backwards and shit, super annoying. Maybe i should be pairing it with another control net at the same time or something.
Ok let's face it - the prompt "masterpiece" does not do ANYTHING and is pure superstition. If not, just let me know what a "masterpiece" is and how the AI would define it - a Mona Lisa? 😂
#### Links from my Video ####
Lighting Based ControlNet civitai.com/models/80536?modelVersionId=85428
Negative Embeddings: docs.google.com/document/d/1ODIUsxT0Am1QX0BDvtUbXYIw7LHFS-TzVxjyAShhQBw/edit?usp=sharing
Install ControlNET: ua-cam.com/video/zrGLEgGFJY4/v-deo.html
#### Join and Support me ####
Buy me a Coffee: www.buymeacoffee.com/oliviotutorials
Join my Facebook Group: facebook.com/groups/theairevolution
Joint my Discord Group: discord.gg/XKAk7GUzAW
Imma need that moon circle
Your content is insane, and the community around AI (both LLMs and image generation ones) is one of the best.
We get so many new things every single day, that it's already impossible to keep up. You have my thanks for making it easier.
Really cool, I've seen Clipdrop has a dyamic relight tool for editing multiple lights based on depthmaps. It would be cool if a tool like that would be integrated in A1111.
that would be awesome :)
I agree, it would be amazing!
Wow, this looks like a fantastic tool to aid composition, many thanks for letting us know about this tool, I'm off to try it now.
It's quite amazing how whenever I think, "Hmm, I need to know about X in Stable Diffusion/ControlNet/LORAs..." and bingo, Olivio has it covered! I was just thinking how it would be easier comping stuff into AI-generated backgrounds if I could get the background lighting to match my lighting setup here in the studio at home. And this looks just the job for that. Your last example is just really, really clever. Shows once again how old-school comping knowledge is the key to a really good AI image. Bravo!
This is nice but I find that controlling lighting is pretty consistent via prompt, like color+ backlight, spotlight rimlight underlight even flashlight works,. also mentioning daytime or weather conditions like , day night dawn dusk overcast sunny hazy foggy dusty etc.
no it doesnt always work
Olivio, you are the best AI image guy on UA-cam.
this is great that it has a model, I've been doing this for months using gradient maps and img to img. Might make a video of that as an alternative method. Great video as usual Olivio thanks.
Woah, cool update. Thanks for sharing, Olivio!
I was already doing this by editing the black and white that softedge creates and inpainting but this is much better!
You sir, are a gentleman and a scholar. Thanks for all the research you do so we don't have to.
It's kind of a nice CN addition, but after playing with it a while, i dont see the point using it. It is way more easy to add some light in photoshop where i have full control over what im doing than trying thousand times to get the right light from the AI, which i need to adjust in PS afterwards anyway. I mean you create a light map in PS, why just dont using this map with your final image in PS ;)
This is really awesome for those QR-Codes I assume👍
Love your result, Nice technique
Been subscribed for a while now and loving your content. Have learned so much more.
If it's even possible to render a car with properly round wheels (I've spend many hours trying), would you be able to do a video showing how it can be done? Keep up the great content.
I love all characters 😂😂 this is how you tell a story with in 2023
amazing thanks for posting this video, was looking for something exactly like that
you are very welcome my friend
WOW! A really cool feature!
Amazing video! Learned a lot as always
It would be cool to change the lighting in real time, such as by dragging different light sources around.
Yes, a lighting editor like the OpenPose editor would be a dream for people like myself who use SD only for inpainting and comping. You'd just add the lighting like in your home setup/studio and then generate a background that is a really close match to your own lighting, making the relight so much easier. Even this already gives me incredible results.
😃Hey love your videos! Was wondering if you might be able to do a step-by-step guide on how to uninstall and do a clean install of auto1111? Something happened to mine and it just stopped launching. Tried to reinstall it and I still have issues with it not launching. Last thing I did was add ROOP and update it. After that everything went bellyup :(
Cool, but doesn't the depth CN model (kinda) already do this?
Thanks!
where is the lightingBasedPicture_v10.yml? can it works without the yml?
I've been experimenting with this for awhile now and find the workflow quite finnicky and limited to be honest. You can get some interesting results for sure but if you'd like to "re-light" any of your older images then you're in for a rough time.
I think a lot of people commenting here hasn't actually tried this out so be warned, it's not quite as easy as it might seem =)
Yep! It's real bad. It's interesting though, and promising as a concept. But it seems basically to be a 100% lock on all transformer layers that you just scale back and kind of hope it moves your image in the right direction. It certainly doesn't do what it says on the tin.
Thank you :)
Control Net in Comfy or Invoke with SDXL ?
Amazing! 👍
Unrelated question/ tutorial idea:
Is it possible to combine the texture from person A with the features of person B?
Examples: a female version of the rock band Kiss or Alice in Wonderland with Alice Cooper's makeup.
Nice!
If you are a concept artist this is the best way to use AI
Perfect
Amazing video!
Thank you so much
Does anybody know whether something similar has ever been produced for SDXL?
오늘도 영상 잘 봤습니다 감사합니다 올리비오
cool
Open pose never works for me. It puts the persons limbs in the right spot but always like backwards and shit, super annoying. Maybe i should be pairing it with another control net at the same time or something.
Doesn't work
Ok let's face it - the prompt "masterpiece" does not do ANYTHING and is pure superstition. If not, just let me know what a "masterpiece" is and how the AI would define it - a Mona Lisa? 😂
Man put the prompt in description to be copied especially negative