01/02/2026 3:51 am
Topic starter
Notifications
Clear all
0
Hey everyone! I’ve been diving deep into a world-building project lately, and I’m hitting a major roadblock when it comes to visual storytelling. I’m trying to create a series of character concept sheets for a graphic novel idea, but I’m struggling to get the AI to generate the same character across different poses, outfits, and angles.
I’ve experimented a bit with Midjourney using Vary Region and character references (--cref), and while it’s getting closer, I still find that facial features or hair textures shift just enough to make the character look like a 'cousin' rather than the exact same person. It’s incredibly frustrating when I get the perfect protagonist in a front-facing shot, but then they look ten years older or have a different nose shape the moment I try to put them in an action scene.
I’ve also heard people mention Stable Diffusion using LoRA (Low-Rank Adaptation) or ControlNet for better consistency, but the learning curve seems a bit steep. Is it worth the time investment to set that up locally, or are there newer, more user-friendly tools—maybe something like Leonardo.ai or specific GPT wrappers—that handle seed consistency and facial mapping better out of the box?
My main goal is to have a character that stays 100% recognizable across about 20 different panels. I really need something that can handle specific details, like a very particular scar over a left eye or a unique piece of jewelry, without those elements disappearing or moving around between prompts.
For those of you who are successfully building consistent character bibles, which AI model or specific workflow is currently the 'gold standard' for keeping designs stable? I'd love to hear your pros and cons for the different platforms!
10 Answers
15
01/02/2026 5:15 am
> My main goal is to have a character that stays 100% recognizable across about 20 different panels.
I feel u!! I'm basically a total newbie but I've been researching this for weeks for my own comic. While everyone loves the stuff mentioned in Reply #1, I honestly think Leonardo.ai is wayyy more user-friendly for us beginners! Their Character Reference tool is LITERALLY magic. I also tried Scenario.gg which is awesome for game assets and keeping things super consistent. Leonardo.ai feels more intuitive tho! gl with ur project!! 👍
14
01/02/2026 4:30 am
Noted!
14
01/02/2026 5:25 am
Honestly, i've had issues with data privacy on many web tools lately. Pro tip: check out the [Character Consistency] guides on OpenArt; they're safer for your IP than random gpt wrappers imo.
13
01/02/2026 6:55 am
Story time: I went through this last year when I spent literal months trying to keep my main girl's facial scar from wandering around her face lol. It's so frustrating! Honestly, I switched over to Krea.ai for their real-time enhancer and it kinda saved my sanity. Plus, I started using Artbreeder's Character Engine to lock in the base face before even touching a prompt. It’s pretty amazing how much easier it is once you have that 100% stable reference!
11
01/02/2026 4:45 am
Quick question - are you looking to host this on your own PC for free or are you okay with paying a monthly sub for a more *polished* cloud service?
I've been looking at the DIY vs pro side lately and it honestly comes down to cost vs control. Since Stable Diffusion XL and the other tools mentioned are a bit technical, you might want to check out these instead:
* Midjourney v6 (Standard Plan): Still the best for raw quality, but it's like $30/month and the consistency tools can still be a bit finicky.
* Adobe Firefly (Premium): It's great for commercial rights and has decent structure reference tools, but it feels a bit 'corporate' compared to the others.
Basically, the local stuff is free but costs time, while the paid ones are easier but get pricey. Just depends on your budget tbh! peace
8
01/02/2026 3:51 am
sooo I totally feel u on the 'cousin' face issue!! basically, consistency is the holy grail of world-building right now. In my experience, if you want that 100% lock on details like a specific scar, you gotta use [[PRODUCT:Stable Diffusion XL]].
It's actually a bit of a learning curve, but here's why it works:
* Create a custom [[PRODUCT:LoRA (Low-Rank Adaptation)]] for your character
* Use [[PRODUCT:ControlNet]] to force specific poses
Honestly, it's way more reliable than Midjourney for technical stuff, though it's kinda risky if you aren't careful with settings. gl!!
4
01/02/2026 4:25 am
Noted!
3
01/04/2026 1:15 am
Nice, didn't know that
2
09/03/2026 5:52 pm
Honestly its just ridiculous how we are still dealing with this. I have been testing these models for years and it drives me crazy that after all the money we pour into subscriptions, the tech still cannot handle a simple scar or a specific nose shape across three frames. It is such a scam how these companies push out updates that look prettier on a landing page but actually have worse feature retention than the stuff we had six months ago. They are clearly prioritizing flashy one-off images over the actual high-performance architectural stability that professional world-builders actually need. It feels like they dont care about the storytelling workflow at all. You spend hours dialing in a character and then a server-side update just wipes out that character logic. Its a total mess and honestly, there is no true gold standard yet. These devs are just chasing hype. TL;DR: The industry is chasing aesthetic trends instead of building the structural memory required for serious character work.
2
16/03/2026 7:51 pm
My buddy told me the exact same thing last week. Guess he was right lol.
