Master High-Res AI Image Generation with Realistic Vision V3.0 + ComfyUi

Hello friends today the new version of The Realistic vision is released the name is version 3. and also it has vae embedded it's very important and the quality is very high I did some tests on realistic vision and you can find my test here in the first page these are my tests and I figure out it has very very good consistency and I like that for architectural design prompting okay now I want to focus and review the new version and also I want to show you how you can use the iris fix method in config UI okay so in config UI now I am going to work with the default notes by the way myself I'm working with efficiency notes but this time I want to show you that how you can use the default notes and make high risk fix and also using the realistic Vision version 3. okay at first I have to load the checkpoint realistic Vision version 3 and also it's very important because it has the vae insight and merge it in the checkpoint okay now I have to write my prompt already I have my prompt from the morning and I want to use that okay let me adjust my resolution and for the sampling DPM plus plus 2m and Karis okay let's start Q and you can see the process the quality and the consistency inside the image is very better and really I like the result mm-hmm yes it's very nice and also you can see my prompt I wrote about the living room care Revolt futuristic Style design lighting Golden Lights pocket pool and maybe this time I have to increase the width of the living room and here again you can see the living room with the pool and I like the consistency because it's very important to place the sailing connective walls and place the furnitures in correct positions with the first try and really it's very cool so uh now uh I want to make a high-risk fix okay and if you go to the realistic vision I will place the link in the description you can find this link okay from this link you will go to the High um hygiene phase and you should download the 4X Ultra sharp they suggested us to work with this method for using in ibisfix okay now what should I or what can I do with confi UI to make the high risk fix and make image in higher resolution okay at first I need another case sampler I prefer to copy from this one okay this is my case sampler I press in Ctrl C and after that Ctrl shift V okay and here you can see that I have the uh wires connected to my inputs it's very important because it will make easy for me it can make a very easy way for me and just I have to connect the latent to latent image but before connecting the latent to latent image I have to obfuscate the latent okay so excuse me I have to connect the latent to latent upper scale by and with this node you can upscale your latent with the ratio for example 1.5 and in realistic Vision they have suggested us to do the upper scale between 1.12 up to the two okay and the denoise string should be about 0.25 up to 0.45 so here is my input and this is the noise excuse me there was an error in recording and one more time I want to explain important parameters to you and after that uh you will see the video excuse me for that but uh take a look it's very important from first case sampler the data will be exported to the vaed code from the vaed code it will be an image the image go to this node upscale image and after that you will load your upscale model after that it's going to Upper scale image if we are resizing the image down and it is going to the vaa encode and it's coming to case sampler again after that from the latent we have to create a vaed code and connect the VA here from this point to this point and add the save image node it's on and continue the video thank you when the image is going to this node with this upper scale method the size is very high and again I am trying to resize the image to lower size and this is the final size that I want and the final image will go to the VA in code and after that it can go to the case sampler and in case sampler it can start the process so I am going to start the process it's finished you can see that it is going to these nodes and after that you can see it here okay great you can see that that this image is in very high quality yes really the quality of the output is very good about the contrast everything everything in this image is very good and you can see that I didn't use any additional vae I didn't use any Laura and this is I think this is the power of the realistic vision and with these upper scale method or high risk fix you can get a great result and in confi UI it is possible to make the high risk fix for several time okay let me show you this time the resolution of this image is uh you can't see let me make my windows smaller okay 1200 in thousand two hundred okay and again I want to make another high-res fix to make a 4K image okay for example I have to select all these nodes mm-hmm let's just select them Ctrl C and Ctrl shift V okay I have them and here from the latent I have to connect the excuse me from my vaed code I have to connect the VA edit code to the upper skate image and after that write the final resolution for example 2048 in 2048. and every other thing could be same and again Q prompt okay it is going to make the first image after that it is going to the upper scale method okay the higher resolution image is started in processing now we have the image in higher resolution 1200 in thousand two hundred again we sent the image to another upper scale method with the size of 2048 and 2048. it's in vaed code when the size uh often when when you are increasing the size of your image after that it will use lots of vram and also lots of RAM on your computer uh totally maybe you can think that 2048 until in 2014 is not a big image but for AI today it's a big size I am sure in the early future everything will be changed and easily we can generate this resolution without any high-res method but now we don't have any choice and we should try to work with this workflow and try to learn the AI and when something comes with easier workflow after that we can use that okay so let me arrange the photos okay from left to right you can see that my screen had I had some lag on my screen okay this was the first image with a lower resolution this is the second image with some higher resolution and here better you can see the textures and some details and here is the highest resolution and in some okay we have that and you can see that after three level of high-res fix we have this image let me open that in browser and this is actual size yes this is actual size without any Zoom and it's very interesting yes it's very very interesting and also a hint you can add one order high-risk fixed level and again you for example four times six times seven times and also there are some method for caching the latent and also they are very useful but they will be some Advance it is some Advanced workflow but in this video I try to explain you how you can use the upper scale method or high risk fix okay it's a high risk fix in comfy UI and how you can use the realistic vision vae with Eva because it has the vae embedded and I want to show you that what is the high quality of the realistic vision it's all uh hope to see you soon with the next video for the next video I want to discuss about the Laura and some other methods like hypernet works um negative embedding or some other tricks that maybe we can use them in architectural workflow uh hope to see you bye

Channel: Arch_AI_3D

Views: 3,354

Keywords: AI, stable diffusion, AI architecture, ComfyUI, ControlNet, architecture, artificialintelligence, Tutorial, ai robots, ai tools, artificial intelligence

Id: 9HYFS97b5QI

Length: 15min 35sec (935 seconds)

Published: Wed Jun 28 2023