My Experience Making Webtoon Backgrounds with AI Art (Practical Notes)
I draw webtoons as a hobby, and because I felt I was wasting too much time drawing backgrounds, I tried using AI images for backgrounds for about half a year. To get straight to the point: "Use it, but be prepared to fix a lot by hand."
1. With prompts, locking in a style is key
If you just throw in "a cute cafe" every time, the tone comes out different each time, creating a jarring feel from panel to panel. I lock the seed value and save color/paint-stroke keywords as a set.
2. It's faster to set the composition yourself
If you just use text like "an alley with perspective," trial and error can take 30 minutes each time. It's much faster to put in a rough sketch and generate based on that.
3. Never delegate people
Fingers, eyes, clothing wrinkles... every single one is terrifying. Backgrounds only.
4. Copyright/Ethics
I'm still thinking about this constantly. I use a paid model and have checked the commercial-use terms, but honestly, I'm scared of how readers would react if they knew. So even though I only use it for backgrounds, I disclose that I used AI in my work notes.
My work speed feels about 40% faster. But the time spent grinding on prompts makes up for it... so total time is cut by about 20%. In the end, a tool is just a tool.
9 answers
Got a source? If there's anything that quantitatively measures how much fixing the seed contributes to maintaining tone, please drop a link.
I didn't know about fixing the seed value — just saved this right away. Thanks!
Fingers lolol you really can't leave this to humans
You really did the right thing by disclosing that you used it. I did the opposite—I hid it at first and kept publishing, and then one day a comment saying "Isn't this AI?" appeared and I completely lost my composure... In the end, I posted a notice and apologized, but it would have been much easier to just be upfront from the start. Readers turned out to be more sensitive about the 'hiding' than about AI itself.
I've also been thinking about the copyright aspect, but even with a paid model, isn't there still some commercial-use risk depending on the source of the training data? I'm glad you checked the terms of service, but since platform policies can change later and sometimes be applied retroactively, I'd recommend keeping backups and usage logs.
Agreed, the background is the real problem... I can manage the characters somehow, but I end up wasting half a day on the background.
Well, would the excuse "I only write backgrounds" really fly? Readers these days can already tell. If it were me, I'd just state it clearly on the cover rather than in the production notes.
Isn't 40% a bit of an exaggeration? If you add up all the time spent writing prompts, tweaking them, and correcting the output, I'd say it's realistically closer to 15–20%. If you've been using it for half a year, do you at least have data from timing each cut?
I really relate to #2. I also wasted an entire day early on trying to nail composition with prompts, but after I switched to roughly sketching in Clip Studio Paint and running img2img, I got it down to under 10 minutes per panel. I think the moment you try to get perspective right with text, you've already lost.