great insight! i think stable diffusion and control has great potential to replace certain rendering engines. i use it almost everyday but the biggest challenge is to fine tune and find the perfect parameters for a really crisp and accurate generation on SD and Controlnet. what model, sampling steps, CFG scale, Loras etc
Hey, thanks a lot for your comment! Absolutely agree, it is amazing for the initial idea visualization, but as you said, I can't say the same for the case of final views with all the details and different materials. I am trying for that hopefully, I will share a video about that case too :) I have used Realistic Vision V2.0, V3.0, and EpicRealism model. Between 24-40 sampling steps, 5-9 CFG scale and I didn't use any additional Loras. Which model or Loras do you use?
@@designinput are you having consistent results with that? i used realistic vision 2.0, 3.0, Xarchitectural, Deliberate, Reliberate, between 25-50 steps. out of all this i get good result from realistic vision 3.0 with 40 steps and CFG scale of 15. lower CFG scale are good but sometimes the image comes out inconsistent with my prompt, like i want the walls to be white but they sometimes come out with a different material, so upping the CFG fixed that for me. as for control net what setting are you using? Loras i am using XSArchi_129, for glass windows and Lora Add-Detail that can somewhat add some crispness to the image. out of all the samplers i found DPM++ SDE Karras the best
Looks great. Do you have a way to use a photo as a background but have a CAD drawing instead of a sketch? Its maybe more complicated as Camera views of models are difficult to match onto a photo. will AI understand the differen viewpoints and still merge them?
Hey, once again amaizng video - so for the interior AI rendering, which packages I need to download (I want to put them on my runpod server and see if it works)
I appreciate how you describe the process in detail. I wonder if you are able to describe certain materials/furniture in stable diffusion to be more accurate or is there any script that can help to do this? And could you also describe a bit your computer setup please? I am interested in implementing it in my work.
Hey, you are very welcome, thanks for the feedback! You can try using multi-control with segmentation for more control but still, you may not be able to describe all the elements in your design. I am planning to share a workflow about that soon, hope that can help!
Hi, it wasn't possible to do that when I shared the Fabrie video. But they just added an inpainting feature, where you can edit the parts of the images. However, I don't think you can insert your own sketches
Çok teşekkür ederim Fırat :) Automatic1111 arayüzünü kullanıyorum ama eğer model olarak soruyorsan, kullandığım bütün modeller genellikle 1.5 base modelli, SDXL'ı bekliyorum :)
Hey, apparently V3 got lots of negative comments that's why the developer uploaded a new V4. I haven't tested it out a lot yet so can't say much about it. But I will share my results as a comparison soon :) Thanks for your comment! What do you think about it? Which one did you like more?
great insight! i think stable diffusion and control has great potential to replace certain rendering engines. i use it almost everyday but the biggest challenge is to fine tune and find the perfect parameters for a really crisp and accurate generation on SD and Controlnet. what model, sampling steps, CFG scale, Loras etc
Hey, thanks a lot for your comment! Absolutely agree, it is amazing for the initial idea visualization, but as you said, I can't say the same for the case of final views with all the details and different materials. I am trying for that hopefully, I will share a video about that case too :)
I have used Realistic Vision V2.0, V3.0, and EpicRealism model. Between 24-40 sampling steps, 5-9 CFG scale and I didn't use any additional Loras. Which model or Loras do you use?
@@designinput are you having consistent results with that? i used realistic vision 2.0, 3.0, Xarchitectural, Deliberate, Reliberate, between 25-50 steps. out of all this i get good result from realistic vision 3.0 with 40 steps and CFG scale of 15. lower CFG scale are good but sometimes the image comes out inconsistent with my prompt, like i want the walls to be white but they sometimes come out with a different material, so upping the CFG fixed that for me.
as for control net what setting are you using?
Loras i am using XSArchi_129, for glass windows and Lora Add-Detail that can somewhat add some crispness to the image.
out of all the samplers i found DPM++ SDE Karras the best
Outstanding! Love it, thank you
Hello, thanks a lot :) So happy to hear that, you are very welcome!
Thank you for sharing this 😍
My pleasure 😊 Happy to hear that you liked it!
Looks great. Do you have a way to use a photo as a background but have a CAD drawing instead of a sketch? Its maybe more complicated as Camera views of models are difficult to match onto a photo. will AI understand the differen viewpoints and still merge them?
Thats amazing I’ll use it on my next university project!!
Thank you! Happy to hear you liked it
I’ve been looking for explaining like this!
Thnx you🙏🏼❤️🩹
You’re welcome 😊
Great video, thanks
Hi Pablo, thank you, glad to hear that! Your are very welcome!
Hey, once again amaizng video - so for the interior AI rendering, which packages I need to download (I want to put them on my runpod server and see if it works)
What do you mean by packages, checkpoints? If yes, you can try Realistic Vision 5.1 or epicRealism ones for SD1.5 version
Great video!
Thank you :)
Awesome
Thank you
I appreciate how you describe the process in detail. I wonder if you are able to describe certain materials/furniture in stable diffusion to be more accurate or is there any script that can help to do this? And could you also describe a bit your computer setup please? I am interested in implementing it in my work.
Hey, you are very welcome, thanks for the feedback! You can try using multi-control with segmentation for more control but still, you may not be able to describe all the elements in your design. I am planning to share a workflow about that soon, hope that can help!
how to make this with Fabrie ? they seem not to have a direct controlnet access ?
Hi, it wasn't possible to do that when I shared the Fabrie video. But they just added an inpainting feature, where you can edit the parts of the images. However, I don't think you can insert your own sketches
@@designinput yeah I saw the change - I have to text it. I work mainly with Vizcom.
Süper video Ömer. stable diffusion 2.0 mi arayüz?
Çok teşekkür ederim Fırat :) Automatic1111 arayüzünü kullanıyorum ama eğer model olarak soruyorsan, kullandığım bütün modeller genellikle 1.5 base modelli, SDXL'ı bekliyorum :)
Great video ! Is stable diffusion free ?
Hey, thank you :) Yes, it is absolutely free to use it locally
What do you think of realistic Vision 4, do you like the features of Vision 3 better?
Hey, apparently V3 got lots of negative comments that's why the developer uploaded a new V4. I haven't tested it out a lot yet so can't say much about it. But I will share my results as a comparison soon :) Thanks for your comment!
What do you think about it? Which one did you like more?
Is it possible in MJ?
Hey, you can do it for the midjourney generated images but not for your own images :/