I didn't pay much attention to the story at first, but when I read that it was the first comic on this site, I reread it from the beginning and have to say, it's not bad.
Now I'm really curious to see how the story continues.

Well this just made my day..

That bird stole my attention

xxxmistercoolxxx said:
I didn't pay much attention to the story at first, but when I read that it was the first comic on this site, I reread it from the beginning and have to say, it's not bad.
Now I'm really curious to see how the story continues.

It’s incredibly hard to do comics in AI since by design its output is random. You can steer the output by using different models and Loras. But to get the same exact caricter every time is witchcraft to me.

furryranch said:
It’s incredibly hard to do comics in AI since by design its output is random. You can steer the output by using different models and Loras. But to get the same exact caricter every time is witchcraft to me.

No pun intended

furryranch said:
It’s incredibly hard to do comics in AI since by design its output is random. You can steer the output by using different models and Loras. But to get the same exact caricter every time is witchcraft to me.

I don't know anything about AI, but from what you wrote, it must be pretty difficult to use.

xxxmistercoolxxx said:
I don't know anything about AI, but from what you wrote, it must be pretty difficult to use.

It's pretty easy actually once you get the hang of it. What to prompt for, etc. But like i said, to get EXACTLY what you want is near impossible without a specific LORA. LORAs require thousands of pictures to train off of. Not hard to get when you have a character like, say, Kat. If it's your own bespoke creation though? You have to either hand draw it thousand of times, or use AI, then edit the pictures to a "close enough" approximation, then use those to train your model.

furryranch said:
It's pretty easy actually once you get the hang of it. What to prompt for, etc. But like i said, to get EXACTLY what you want is near impossible without a specific LORA. LORAs require thousands of pictures to train off of. Not hard to get when you have a character like, say, Kat. If it's your own bespoke creation though? You have to either hand draw it thousand of times, or use AI, then edit the pictures to a "close enough" approximation, then use those to train your model.

Nearly spot on!
For most of the published pages 70% of the consistency came from prompting the right characteristics maybe 5-10% from some rough lora that I made for Tom and Jack and maybe 10% from manual drawing/sketching + inpaint (for exemple I have to "re-draw" Jack horns on almost every panel cause the model I used did not like this kind of curled/ram horns so I had to "guide it")
Then there is always 10-15% of inconsistancies that I have to consider acceptable (otherwise the time I'd spend in editing would make my progress prohibitively slow)

Right now I'm training some lora for illustrious (for the next chapter), the first draft for Tom and Jack is rather encouraging (I used ~50 cleaned up images for each).
I hope to improve it a bit with the second batch (better images + better compatibility with other lora + smaller size) and then do the other characters 🤞
If I succeed, this should increase the quality and consistency tremendously.