Mastering ComfyUI: How to Use Embedding, LoRa and Hypernetworks! - TUTORIAL

hello everyone and welcome back to Dreaming AI my name is nuked and today we are going to learn how to use embedding Laura and Hyper Network income V UI embedding also known as textual inversion Laura and hiker Network are an alternative way to control the style of your images and stable diffusion it's like fine-tuning the model itself but in a separate file for example we have them for a specific type of eye drawing style or even a person I won't delve into the technical details of these models as there are videos out there that explain them much better than I could with more expertise instead I'll show you the Practical use of each of these fine tuning techniques you can find many of them ready to use on civitai.com all you need to do is download them and copy them into their respective folders within comfy ui's models Fuller um for this tutorial I've downloaded very bad image negative for embeddings add more details for Laura and Louisa P pixel art for hybrid networks for this tutorial we'll use a workflow that I've prepared divided in two simple parts for image generation in one we'll apply the additional models while in the other we'll leave it unchanged to ensure the results are comparable both parts share both the empty latent image and the seed you can find the download link in the description okay so let's start with embeddings using embeddings in comfy UI is done by invoking them in the text prompt just like an automatic 11 11. but with a different syntax in comfy UI we need to use an open parenthesis in Bedding a Colm the name of the embedding file we copied to the folder another colon and a numeric value representing the strength with which this embedding will be applied to the image typically values ranging from zero to one are used the higher the value the more visible the modification from the embedding will be in the resulting image of course there are also embeddings designed to be placed in the make it a prompt to used to remove something instead of adding it let's apply very bad image negative to the negative prompt and see the result with and without it foreign well I'd feel the difference is is quite evident you can of course use multiple inventing simultaneously now let's move on to Laura Laura stands for a low rank adaptation and similarly to embeddings applies a modification to the model's output speaking of preferences many people prefer to use Laura because it seems to have a more impactful and consistent effect on the output to use a Lora model we have a specific node called Laura loader which shows us the list of Laureate detected in the folders scanned by compui it takes both the clip and the model from our Checkmate loader as input and Returns the same but fine-tuned so let's select our Laura in the loader in my case add details uh just like with embeddings but a bit more annoying you can use multiple Lora models together to do this you will need to stack the lower loaders one after the other also the lore loader has two parameters that you can adjust to regulate the intensity of the lore's influence on clip and the model and therefore the result uh as far as I know there isn't a precise rule on how to use these parameters but I recommend testing them as you go to achieve results closest to your expectations like embeddings let's try a test with and without to understand how influential they are in the output however let's align the negative prompts so that they're the same in both images now let's generate the image and see the result foreign let's try with different intensities I'd say the difference is perfectly visible finally let's move on to hyper Networks uh hyper networks are probably the oldest technique among those mentioned conceived by the developers of Novel AI but in the last few months it seems to have been somewhat neglected the application of a hypernetwork is very similar to that of Laura we have a specific component called hypernet workloader where we should input our model which will then be returned with fine tuning applied the model output like in the case of Laura should then be connected to the usual sampler and so let's try Louisa pixel art and see the result foreign well I'd say the pixel style has been applied quite well although the image is somewhat different that's all for today please consider liking and subscribing if you found this tutorial useful um also if you have any questions please let me know in the comments below I'll be happy to help you out as much as I can and until next time heat dreaming [Music]

Channel: DreamingAI

Views: 999

Keywords: LoRA, embedding, hypernetwork, finetuning, image generation webui, anime, ComfyUI, programming, python, advanced, text generation, AI, stable diffusion, artificial intellingence, dreamingai, ai news, best free ai, best ai model, dreamingai tutorials

Id: PrHST5kGAFo

Length: 7min 27sec (447 seconds)

Published: Fri Sep 22 2023