StableDiffusion XL Masterclass: Custom trained AI in production

A Masterclass in concepting with Stability AI's SDXL
using Automatic1111 with custom LoRAs
(did that sentence scare you? Read on brave one)

• Yes, I use a well known shoe brand as an example of creating "likeness", but it extends to both the likeness of a product or the likeness of a style!
• The software we use is FREE forever (Adobe is about to start charging extra for generative fill - get out of here!)
• Beginner friendly - From zero to hero without any technical stuff getting in the way
• Nearly 4 hours of guides, workflows, tips and tricks from an established 3D artists who use this in production right now
• Save tons of time when doing rough concepts - wow clients and get to "final_final_final" faster and with better visual aid along the way
• Easy and fast training of your own digital art, photography, 3D or any other image to get your unique aesthetics and look - no stock here!
• You can grab my own custom trained SDXL model here - highest quality available and downloaded over 13K times with 5⭐/5
• SDXL Lora training presets for Kohya - for the highest possible quality on consumer GPUs; load it up and train instantly!
• Use offline and locally on minimum 8 GB VRAM or connect to an online GPU - the latter means no hardware required except a laptop!
• Kick Midjourney in the butt and come to the light side of free offline hugs, great UI and high resolution rainbows!
• Windows, Mac or Linux agnostic! (We will be using Windows 10)
  

Course Overview:

In this course I will teach you everything you need to know to start creating concepts and images or elevating existing ones with the aid of generative AI using SDXL models from StabilityAI (copyright free and "clean" models) enhanced by LoRA models trained on your on art/images. It is aimed at artists working as 3D artists, Painters, Photographers, Concept Artists, Character Designers, Pattern Creators, Fashion Designers or anyone who does sketching for projects. The course is comprehensive; talking about copyright, installing all the software and custom models and extensions, getting up and running - without all the technical jargon - and then showing you how to train an SDXL Lora on your own art for a fast and lightweight little model (~800mb) that you can use forever and for infinite use cases. Is essentially a model of your DNA as an artist for you to use however you want!

Get ahead of the curve!

We dive into creating concept art for a motion graphics piece of a fictional sneaker shoe ad. I will show you free open-source software for upscaling your existing images and getting our dataset ready for the training process - dead easy. I then show you how to start the training with the provided presets, making the training effectively plug and play - all you need to do is provide the correct folder paths and off you go! We then take the LoRA we just trained into our Automatic1111 WebUI and start churning out very high quality concept pictures to impress and inspire our fictive client. I also show how to create endless variations without changing angles or backgrounds, and how to implement client changes without changing what the client already approved. Handy dandy!

We finish of our concept art where I show you my custom workflow for advanced upscaling and creating even more details in our images. This workflow can in fact upscale your images to infinity if you have the time to wait for the render!

I come from an artist background so that is my lingo, and I assume you have basic knowledge of Photoshop or any other creative software and know your way around a computer (even if it's just looking up videos on YouTube).

FAQ:

1. Why get this when Adobe is implementing AI in all their products as of October 2023?
   
   Adobe will start to charge extra for their AI stuff starting November 2023 with "generative credits". In fact, many of the technologies they have implemented are built upon the same tech we use in the course. Their models are pretty weak and trained on generic stock photos. The difference is once you've learned my workflow you will never need to pay them an extra cent again (except your subscription of course, I use Affinity Photo). In addition, there are even clever Photoshop Plugins (for free) that can implement a custom generative fill once you've learned to use Automatic1111 with my course. This workflow is insanely powerful - sort of like a custom Generative Fill trained on your images!
2. What moster of a computer do I need?
   
   You need a computer running Windows, Mac or Linux and that has a GPU with at least 8GB VRAM. Your workflow will be slower but it will totally work. Train over night is my answer to that! Recommended is 12GB VRAM for the smoothest ride or rent-a-GPU* for a couple of cents an hour! In addition to this, I'll be providing you with my settings files so you can take advantage of all the optimizations I've already unlocked!
3. I'm an Artist, I don't have a PhD in Computer Nerdery!!
   
   That is totally fine, I'll hold your hand through all the steps of installing the software + models and there is a whole chapter on using the interface. I explain all the common pitfalls of using SDXL as well as cover all basics of generating your first pictures. We will go through a shallow theory layer and I even explain it with drawings - yum!
4. Can I sell any images I make with this or will I go to hell?
   
   Absolutely, no hell today! Although the course does assume you have copyright over the images you are training on. In the course we use a well known sneaker brand because that is our fictitious client. I'll provide the Dataset as well since it is CC0 (free for any use). I use the base SDXL-model from Stability AI as the root-model for my model which we then train our LoRA on! See this for more info on selling AI art!
5. OK, I'm excited and ready, can I download the course and store it forever or is it just a rent thing with hidden self destruct?
   
   You own the videos for all time, put them somewhere and use them as a guide if you ever get lost - I'm always there, in the darkness of your hard-drive, waiting to come to the rescue in your AI adventures!
   
   

LET'S GO!

Disclaimer: This artificial intelligence model ("the Model") is developed solely for artistic expression and technological demonstration, not for creating or distributing illicit, infringed or inappropriate content without the consent of the copyright and trademark owner (Nike Corporation). The creator expressly disapproves of such use. The Model is intended to be used within a studio environment for the sole purpose of expediting the creative process within an advertising/design production pipeline. By downloading this course, you ("the User"/"the Student") agree not to misuse it in any manner contrary to these intentions and accept all responsibility for compliance with these terms.

Also,

I am not associated with Nike© Corporation ("Their/They/Them"). Their product is used only for educational purposes under fair use. All free datasets used are licensed from Kaggle© with Creative Commons CC0.
© All trademarks, service marks and trade names (e.g., the NIKE name and the Swoosh design) are owned, registered and/or licensed by NIKE.

They are solely used as a fictitious client in this design course.

SDXL is an open-source image-diffusion model created by Stability AI. Please refer to https://github.com/Stability-AI/generative-models/blob/main/model_licenses/LICENSE-SDXL1.0 for the license of the underlying "SDXL-model" that the model in this course and the models subsequently created with the knowledge presented in this course is licensed under.