Civitai with Stable Diffusion Automatic 1111 (Checkpoint, LoRa Tutorial)

foreign welcome back to the channel before I start with the intro I want you to check out these images [Music] how about this one realistic ain't it maybe your favorite celebrity or your favorite superhero if you are into architecture all these images are created via stable diffusion running locally on my PC without me spending a single extra Penny stable diffusion is an open source model which means many creators are playing around with it and generating new models if you are not getting these quality images you are not correctly using the local install of automatic double one double one by correctly I mean using its full potential so in this video I will explain essential extensions and settings needed to be ready for using Civic AI models second I will explain the different Civic AI models like checkpoints Laura textual inversion Etc then I will explain how to use Civic Ai and import these models into your local stable diffusion with an automatic double on double one install and lastly some tips and tricks on prompting and an easier way to learn the model prompts by the PNG info feature on stable diffusion so let's get started firstly you have to follow my tutorial in the previous video and install stable diffusion up to version 2. secondly follow the procedure to install the ultimate SD extension to upscale images I will link both videos in the description below we useformer to optimize image generation and reduce vram usage if you don't have formers installed on automatic double one double one you need to install it first in order to install go to your stable diffusion folder and right click on web UI user.bed then click on edit in the notepad check out where it is written command line args equals type this exformers reinstalls formers in the format shown on the screen save and close the notepad click web UI and launch stable diffusion [Music] you can see it is installing xformers if you get a message on a new pip version available I will show you how to update the PIP version once everything is completed close this terminal to update the PIP version go to your v e n v folder right click and open a terminal over here type python.exe Dash mpip install dash dash upgrade pip once installed close the window and go back to the main folder now right click on web UI user.bat again and edit delete only dash dash reinstall desks formers as shown save and close now let's understand Civic AI models the first is checkpoint these are base models and usually average two to six gigabytes in size these are also called dream Booth models you have to download them in the models stable diffusion folder directory the second is textual inversions these require a checkpoint model to run I will show you examples of each one on the civit AI website textual inversions are small compared to checkpoint files and they go in the embeddings directory hypernetworks go in the model slash hypernetworks directory when you download a hyper Network you will get an input field in the settings of the web UI you can adjust the strength slider in settings Laura needs to go into the model slash Laura folder lycorus requires an extension to be installed and the downloads go in the same folder as Laura model slash Laura wildcards also require an extension they go inside extension slash stable diffusion web UI wildcards slash wild cards folder textual inversions hypernetworks Laura lycorus and wild cards are all designed to be used with some checkpoint base model I will first install the required extension and then proceed to the Civic AI website so we need to install just two extensions like chorus and wildcards the link to those will be given in the description below so we go to extensions install from the URL and then restart the UI [Music] thank you now let's move on to civitai.com the first thing I have done already is that in the browsing mode I selected safe so no not safe images are shown here I have selected some models and added them to my like models I have added all types of models to show you as an example I will showcase some of them and show you how to use them to generate images let's scroll back up and check for a checkpoint model called dreamshaper you can see a variety of styles and images can be created using this model to use this model we download the 1.99 GB file to the correct folder after the download is complete go back to stable diffusion and click on the refresh icon under the stable diffusion checkpoint now select the model let's go back to civit Ai and browse for some images let's say you want to create this image locally now instead of copying props and negative props and settings like clip skip seed sampler Etc just right click the image and save it in a folder now we go back to stable diffusion here go to PNG info and click on click to upload and upload the same image we just downloaded on the right you can see all the parameters and exif comments are showing all we have to do now is click on the send to text to image all the settings and parameters are here we hit generate there is an exception error that says could not find upscaler so what we do is we copy the upscaler name and Google search it we get the hugging face link you can see it is an essogen upscaler let's click on that and download it we should save it in models s again [Music] after the download is completed restart stable diffusion once done go to PNG info and upload the download image again and click on send to text to image click on generate [Music] foreign now we play around with the prompt I am just making minor changes and you should be careful and the prompt is training for this model let's change the portrait of an Indian girl to an American girl with pink hair keeping the seed exact these images go to the output folder in stable diffusion you can also save and download it to a new location if you wish let's play around with the prompt further and this time remove the seed and keep it -1 for randomness [Music] [Music] let's change the prompt a little more let's say at the beach night Moon moonlight [Music] thank you I messed up somewhere and there is a sun not a moon let me recheck the prompt okay found it sunlight passing through foliage we'll change it to Moonlight and let's change the hair color to blonde [Music] [Music] perfect and we got the moan as well let's go back to civitai.com [Music] browsing for some other image the sports card looks nice procedure as last time foreign [Music] let's just hit generate first without changing anything [Music] [Music] let's change it to a Lamborghini [Music] foreign and delete this high quality Road and replacing it with Iceland Northern Light foreign as well and put night beautiful [Music] let's add ice Lagoon black Beach the location is more specific and blue Lamborghini [Music] foreign accurate location and color let's try randomizing the seed foreign [Music] nice increasing steps to say 60. curious to see what that does to the image [Music] very nice I hope you get the picture of how easy this is and it's free [Music] let's move on to another model let's return to civitai.com and check out Alora model using a checkpoint base foreign this is a 3D rendering style Laura model but uses riff animated checkpoint [Music] I will first download the Lora model and then the Rev animated model [Music] [Music] foreign [Music] like before I will save two images and get all the details and settings from it sometimes the model uses an upscaler but does not tell you on the website and sometimes it uses control knit and even that is not mentioned on the website so the easier way is to save the image and check what is missing specifically taking bad girl to see if I can generate Batman [Music] now this image is using esnd that's adenoid C Delta value of 31 337. it's overriding the default value but after doing research it is probably best to set this value for all image Generations somehow it gives better results once I set it the override for that value will go and only clip skip value 2 will remain that's fine clip skips should normally be set at value 1. after changing it in the setting you should reload the UI [Music] another error this image prompt requires a 4X Ultra sharp upscaler we would have never known this if we just copied and pasted the settings given on the website so what I will do now is Google 4X Ultra Sharp upscale.pth upscalers are always in that pth format [Music] anytime you save and upscale you have to close the command prompt and restart stable diffusion [Music] hopefully everything should work now changing the sweater color to Yellow from Red [Music] nice let's see if I can change the background to Green it would go well with the yellow [Music] perfect let's try the bat girl image [Music] thank you [Music] foreign notice how the prompt is bad underscore woman so if you put Batman in one word here it won't work these models of Civic AI are trained with specific prompts so here I should put bad underscore man and it should work [Music] and it worked let me try poison ivy [Music] thank you [Music] to conclude I hope this video tutorial was helpful if you have any questions please ask in the comments the PNG info method is easy and works great always save the image go to PNG info and check out everything about the image this is one of the reasons why many cannot produce the same results shown in Civic AI models and find it difficult I wanted to make it easier for you to be able to create amazing images with stable diffusion just remember some of the models are extremely heavy and require a lot of vram for those getting time out errors try reducing the upscale resolution and use ultimate SD upscale to upscale 2x at a time so from comic books to anime realistic portraits to landscape science fiction macro photography gaming assets you name it there is a free model for everything you may want to do on stable diffusion locally you just need good hardware for it and I just gave you the technical know-how also I have a zip Link in the description below with 50 images which you can import and play around with via the PNG info method as always until next time [Music] if you have watched till here thanks please like And subscribe and hit the Bell notification to receive a notification as soon as we upload a new video [Music]

Channel: ControlAltAI

Views: 11,740

Keywords: stable diffusion, civitai, stable diffusion tutorial, ai art, ai, ai art generator, stable diffusion ai, generative ai, stable diffusion ai art, stable diffusion art, ai generated art, civitai stable diffusion, civitai lora, ai art models, stable diffusion better models, civitai tutorial, stable diffusion lora, stable diffusion civitai, automatic1111, automatic 1111, automatic1111 stable diffusion, a1111, how to install civitai models, stable diffusion civitai extension

Id: 1Ef_nu7DGwk

Length: 22min 40sec (1360 seconds)

Published: Fri Jul 14 2023