Check in

Hello!

I wanted to write up a little document to let people know what I'm currently doing, what I want to improve, and where I think I'm going with this.

I never thought I would keep doing this. This all started as a side hobby but I got hooked. Not sure if its just the process of everything (searching images, preparing the data, etc) or the feedback I got from people. Either way I'm interested in continuing uploading models and how to make this process more efficient.

So below will be a list of some upcoming projects I'm starting to work on.

Document Process

I've been asked about this a few times. There is some hesitant on my side about writing this. Not because I don't want to share, but because my process is mostly following videos and articles that have been shared out. I'll link them here below. However, I think I can list out my current process and tools used. With that I hope others can learn from this, tweak, and improve their own. Also it would be good to hear feedback from others on how they create their models and see what else I can learn.

So I'll be gathering up info for that.

Model Priority Queue

Currently my process for which person/character to model next is mostly me thinking "Hey that person might be cool to model". So I'm always open to suggestions on what people here want. A few of the modeling I've done is feedback from people.

I use Trello to help track model status, from upcoming, processed, trained, extraction, etc. I'm thinking of exposing this so people can know what is coming up.

Along with that I'd like to leverage my buy me a coffee to create a priority order for model creation. If someone is willing to chip in a few bucks then I will prioritize that model way to the top.

I'll be working on a way to expose my upcoming work, so be on the lookout for that.

Textual Inversion/LyCORIS focus

Textual Inversion training started from me only because it was the first thing I learned how to do :). I had trouble starting off with Lora's, but with Textual Inversion I was able to quickly get something going. However, I there are limits to it. When training it on people you can get some really good results, but it can only collect so much data.

When seeing some of the LyCORIS models I was really impressed with how well they can look. More so then some of the Loras I've seen.

So I'm going to keep going with both. If there is some desire to export these to Lora I might think about it if I get enough feedback.

Character creation

I started with the Mix series as a way to try out creating custom characters. I was happy how that went so I will be working on a Mix 2 series. Again I'll pick, at random, about 4 characters to mix and match images with. I'm ready to see what new characters will look like.

Along with that I'd like to branch outside the whole "make a model from a person". Anime, cartoon, etc characters seem like the next option. I'd like to see what that process would be like. Nothing is set now, but dont be surprised if you see some more character creations in the future.

Process improvements

TLDR of these model training is

Find a bunch of images of the person
Crop, resize, setup caption process
Adjust all captions notes
Train
Publish

During those steps I know there are areas I can just automate the process. Finding images, and writing the right captions will most likely still be hand crafted. However the training, extraction, etc I think I can write up some scripts to just move the process. Like most people I'm still using Automatic1111 to help with training. Nothing is wrong with that but once you have a process down I dont really need to load up this giant UI. So I want to focus on how I can run these off the command line.

Along with that I need to take some time to batch run creation of images. As most of you have seen I have a list of similar prompts I use all the time when uploading. Process is run prompt, get about 4-6 images, find the ones that look good then upscale. I'd rather just batch process all that, upscale everything, then pick the ones I like later on. I know Automatic1111 has some ways of doing this with prompt from text file. But I need to figure out how to make this flexible since I like to test with different colors, items, checkpoints, etc.

There will probably be more areas I can look to improve but I'd like to focus on those first.

Thank you!

Thank you everyone who left suggestions, comments, likes, downloads, all of it! I do read the comments and do my best to respond when I can. Seeing people use my work and post images to this site is a pretty cool feeling. I'm glad my work is helping people create their art. So again, thank you.