boobacious

[Announcement] Contagious Augmentation - Part 4

Published: December 11th 2024, 5:25:44 pm

Behind the 11th door of the Advent calendar lies Part 4 of Contagious Augmentation — or well, at least the announcement of it.
But next week, you’ll definitely find the completed comic behind one of the doors ^^

In this comic, we’ll witness more of Alina’s transformation as the virus continues to spread through her body. But we might also catch a glimpse of what’s become of Lisa (the main character from the first two parts) and find out what actually happens when a man is infected with the virus...

The comic should be ready early next week =D

Tasty Temptation - Part 1

Alright, what is there to say about Tasty Temptation? Hmm...

Ah yes! With this comic, I've once again tried out a completely new concept: Weight gain. Though, I should note that it’s not exactly a classic weight gain comic (compared to what you’d typically find online under that term). The protagonist, Mia, doesn’t actually become overweight/fat but instead increasingly becomes more and more pregnant-looking — simply because the latter is much hotter imo and in this fantasy world, I fortunately get to dictate the laws of nature, including how weight gain works (if only that were the case in real life (˘⌣˘) ).
So yeah, in the end, the comic became a sort of fusion between the weight gain genre and my personal tastes (see what I did there? Hehe).

Another fact — or more of a “fun fact” — that didn’t even occur to me until I was illustrating the pages: this comic really has a lot of dialogue compared to my other comics. This wasn’t intentional; it just happened that way. But honestly, I think it fits the comic pretty well.

Speaking of dialogue, I might as well use this opportunity to give you some insights into my process. Before I generate the first image for a comic, I write the entire story. That means the plot, dialogue, character interactions, etc. This is quite a shift compared to how I did things at the very beginning of my “career”, where I would generate images, pick the best ones, and then craft a story around them. Back then, the story followed the images, which made generating them super easy (you had literally unlimited freedom and could keep any cool-looking picture) but the quality and variety of the story and dialogues suffered quite a bit.

That’s why I switched it around: now the images follow the story, meaning the story comes first. This approach makes generating images significantly more challenging — especially when I come up with a cool scene while writing, only to realize later how hard it is to pull off with Stable Diffusion ಠ_ಠ. Sometimes, I even have to (sadly) discard really great images if they don’t fit the prewritten story. That said, there have been a few instances where I was so impressed by a random generation that I rewrote parts of the story just to include it.

To wrap up this topic: while the images are AI-generated, the story and dialogue are not. I’ve tried using AI for this a few times, but it always results in absolute nonsense. Especially since you can’t use flagship models like ChatGPT for it since they filter out NSFW content. I’m telling you this because, in this comic, there was actually one small exception to this rule: the huge order on pages 13/14 didn’t come from me but from ChatGPT. You really don’t want to know what the prompt was haha ^^

Alright, let’s finish with a word about the biggest challenges in this comic. Those were probably the cheerleader costume and the nighttime scenes.

Let’s start with the cheerleader costume. Basically, this highlights the same problem you always encounter with Stable Diffusion when you try to generate anything other than a basic outfit like a T-shirt and pants: consistency. Stable Diffusion can do extravagant, crazy clothing styles very well, but the more complex the outfit, the more variation you’ll see between images, making it hard to maintain consistency across multiple panels. You can often see this in works by other creators who frequently give their characters more elaborate outfits: the individual generations look great, but from image to image, a significant part of the outfit always changes. It has its charm, but personally, I value consistency a bit more.

That’s why there’s a sort of invisible limit on how complex a character’s outfit can be in my comics. Of course, I’m always trying to push that limit higher with better techniques, which I think I managed to do with the cheerleader outfit in this comic. My most ambitious attempt in this area was actually the Halloween comic I posted on October 31st. It was incredibly difficult to make sure the witch’s outfit stayed the same across all the images. (It took me about 6 hours to make this one page Halloween comic :/ ).

As for the nighttime scenes (in this case on pages 18-19): I suspect this is mainly due to a lack of training data, but Stable Diffusion strongly resists generating nighttime scenes. You really have to push it hard with aggressive parameters to force it to create a night scene — and even then, it only works a fraction of the time. If you’ve ever wondered why most of my comics take place during the day, now you know! But yeah, I think I managed to pull off the night scenes in this comic quite well in the end.

Oh wow, this turned out really long. I hope I didn’t bore you ^^
Next time, there’ll be less text and more comic haha.

Until then, take care!