Hello there!
For a brief introduction: me & my partner in life decided we’re going to try making our own adult VN game. I’m really a fan of these games in particular, even if more on the vanilla side. Since our artistic background and skillset is somewhere close to zero, and our true strength lies in unlimited fantasy and storytelling, I decided we’ll embrace some AI tools to help us on our journey.
Boy oh boy, I should’ve known what rabbit hole I was getting into.
I’m about 3 months in now. My background is around 10 years of backend programming, so it wasn’t really difficult to get into the technical stuff. I settled on Stable Diffusion, since I guess there’s no better option for this kind of project at the moment. I have a strong-spec PC, so I can run most workflows locally.
Currently, I’m using stable-diffusion-webui-reForge with the waiANINSFWPONYXL_v130 checkpoint. I’ve “crafted” an art style using about 5 stylistic LoRAs. I have multiple LoRAs for various situations. I embraced inpainting, ControlNet (mostly OpenPose), ADetailer, scaling... basically, whenever a problem came up, I researched how to eliminate it — and the options are endless. I even managed to train some of my own LoRAs to help maintain character consistency. I’m not sure if I did a good job, but I managed to do it — so that’s something.
So, why am I making this post? Since I’m handling the technical side and my partner is more focused on the storytelling, I’ve now reached a point where I’m not sure how to continue. I have two protagonists, and I’m adding some side characters — but I’ve started noticing that those side characters are way out of the art style I thought I had “locked in.” In my mind, I assumed that when I set weights on my stylistic LoRAs and blend the output art style, that would be enough… but now it seems the weights are more like suggestions — they don’t necessarily have a strong impact.
Use case: My two protagonists are young. Since they’re somewhat generic (young, attractive, fit, etc.), their generations come out in a relatively consistent art style — or at least close enough. The problem comes up with other characters, like an old taxi driver or a flight attendant — the art style shifts completely. My best guess is that it defaults to other LoRAs if the “main” ones don’t have relevant material, resulting in a totally different style and breaking immersion. No matter how much I try to “lock” the prompt, or how heavily I weight stylistic cues like “semi-realistic” or “cartoon style,” it still produces a mismatched style.
I’ve come to the conclusion that I either have to redo everything from scratch using some basic checkpoint with one or max two LoRAs — or maybe try switching tools away from stable-diffusion-webui-reForge. I did have ComfyUI set up from the beginning, but since I was a beginner, I went with reForge instead — it felt more intuitive and had a better learning curve. I’m attaching a photo from our WIP game to illustrate what I’m talking about.
You can see that the characters in the center and on the left are kinda close in style (not exactly, but that’s something I could work on). But the character on the right — the old taxi driver — is completely off, and it instantly feels wrong. We stopped at this point, because continuing like this just doesn’t feel right. The result would be messy and chaotic. We want a nice, cozy, enjoyable game — not a hot pile of garbage that doesn’t make sense.
So here we are, at this crossroads — what’s next? What would you suggest? Would you switch tools? If so, is there anything in particular I should try? Our biggest priority is consistency — if we create a character, we need to be able to reuse them. It doesn’t always have to be 1:1, but it has to be close. And then, to keep the art style unified. I know these are basically the number one issues with AI-generated content, but I bet many of you could point us in the right direction. I’ve seen and played a lot of AI art–based games — some of them are really cool. We might not be that cool, but we’d love to at least create something of our own, since we’re truly passionate about this.
I kinda feel like this post is a bit rambly — I hope it makes sense. I’d appreciate any knowledge, experience, or ideas.
Cheers,
M + B
For a brief introduction: me & my partner in life decided we’re going to try making our own adult VN game. I’m really a fan of these games in particular, even if more on the vanilla side. Since our artistic background and skillset is somewhere close to zero, and our true strength lies in unlimited fantasy and storytelling, I decided we’ll embrace some AI tools to help us on our journey.
Boy oh boy, I should’ve known what rabbit hole I was getting into.
I’m about 3 months in now. My background is around 10 years of backend programming, so it wasn’t really difficult to get into the technical stuff. I settled on Stable Diffusion, since I guess there’s no better option for this kind of project at the moment. I have a strong-spec PC, so I can run most workflows locally.
Currently, I’m using stable-diffusion-webui-reForge with the waiANINSFWPONYXL_v130 checkpoint. I’ve “crafted” an art style using about 5 stylistic LoRAs. I have multiple LoRAs for various situations. I embraced inpainting, ControlNet (mostly OpenPose), ADetailer, scaling... basically, whenever a problem came up, I researched how to eliminate it — and the options are endless. I even managed to train some of my own LoRAs to help maintain character consistency. I’m not sure if I did a good job, but I managed to do it — so that’s something.
So, why am I making this post? Since I’m handling the technical side and my partner is more focused on the storytelling, I’ve now reached a point where I’m not sure how to continue. I have two protagonists, and I’m adding some side characters — but I’ve started noticing that those side characters are way out of the art style I thought I had “locked in.” In my mind, I assumed that when I set weights on my stylistic LoRAs and blend the output art style, that would be enough… but now it seems the weights are more like suggestions — they don’t necessarily have a strong impact.
Use case: My two protagonists are young. Since they’re somewhat generic (young, attractive, fit, etc.), their generations come out in a relatively consistent art style — or at least close enough. The problem comes up with other characters, like an old taxi driver or a flight attendant — the art style shifts completely. My best guess is that it defaults to other LoRAs if the “main” ones don’t have relevant material, resulting in a totally different style and breaking immersion. No matter how much I try to “lock” the prompt, or how heavily I weight stylistic cues like “semi-realistic” or “cartoon style,” it still produces a mismatched style.
I’ve come to the conclusion that I either have to redo everything from scratch using some basic checkpoint with one or max two LoRAs — or maybe try switching tools away from stable-diffusion-webui-reForge. I did have ComfyUI set up from the beginning, but since I was a beginner, I went with reForge instead — it felt more intuitive and had a better learning curve. I’m attaching a photo from our WIP game to illustrate what I’m talking about.
You can see that the characters in the center and on the left are kinda close in style (not exactly, but that’s something I could work on). But the character on the right — the old taxi driver — is completely off, and it instantly feels wrong. We stopped at this point, because continuing like this just doesn’t feel right. The result would be messy and chaotic. We want a nice, cozy, enjoyable game — not a hot pile of garbage that doesn’t make sense.
So here we are, at this crossroads — what’s next? What would you suggest? Would you switch tools? If so, is there anything in particular I should try? Our biggest priority is consistency — if we create a character, we need to be able to reuse them. It doesn’t always have to be 1:1, but it has to be close. And then, to keep the art style unified. I know these are basically the number one issues with AI-generated content, but I bet many of you could point us in the right direction. I’ve seen and played a lot of AI art–based games — some of them are really cool. We might not be that cool, but we’d love to at least create something of our own, since we’re truly passionate about this.
I kinda feel like this post is a bit rambly — I hope it makes sense. I’d appreciate any knowledge, experience, or ideas.
Cheers,
M + B
