Even if I end up upsetting some fans, I still have to say it: the "tri." series of Digimon movies and the new 02 movie that followed are just... awful. They're puppets dressed up in familiar names, made purely to cash in. I could rant about them for three days and nights non-stop. What made me even angrier was the new 02 movie—why on earth is Daisuke pale-skinned now? Why is Daisuke pale-skinned? Why is Daisuke pale-skinned? Why is Daisuke pale-skinned?! Thankfully, the character design is so off that it doesn’t even feel like Daisuke anymore—definitely not the bright little sun who embodied every perfect trait.
I’ve been planning this LoRA for a long time. It’s based on my personal favorite Digimon movie—tight storytelling, flawless art style. At first, I was going to make it just about Daisuke, but I’ve always had a soft spot for senpai–kouhai relationship, and I really love the ships “Daisuke × Taichi” and “Koushirou × Taichi.” So I figured out a way to include them all. The outfits are based on the original designs, though generating images with multiple characters sometimes takes a few retries.
The movie’s story takes place at night, so I’ve done my best to reduce the impact of dark scenes, but the output still tends to favor more shadows. You can tweak this using light/shadow-related prompt words. Also, the expressions in the movie are mostly surprised, worried, frowning, angry—almost no calm expressions. So during generation, the eyebrows might come out a raised, and sometimes the lips may look weird, like they have lipstick on. To prevent that, try adding these to your negative prompts:
“pouty lips, puffy lips, thick lips, raised inner eyebrows, furrowed brow, frown, raised eyebrows,”
Because the source material for Koushirou and Taichi is more limited, their generation quality isn’t as good as Daisuke’s, though I’ve tried to improve it as much as I can. I’ve also removed the old film grain feel from the original footage—if you want a more retro vibe, you can intentionally add it back using these in your prompt:
“film grain, vhs artifacts, retro artstyle,”
Below are the trigger tags:
本宮大輔： ayimotomdai1suke, daisuke-brown hair, daisuke-brown eyes,
daisuke goggles, goggles on head,
daisuke blue layered sleeves hoodie, daisuke dark blue shorts, daisuke blue shoes,
泉光子郎： imuzikoushi4rou, koushirou-brown hair, koushirou-black eyes,
koushirou orange raglan sleeves two-tone shirt,
八神太一： imagaytai1chi, taichi-brown hair, taichi-brown eyes,
taichi blue layered sleeves shirt,
If the poses and expressions come out too stiff, try lowering the LoRA weight to around 0.6–0.7, or try increasing the weight of the relevant prompt words. The usual recommended setting is 0.8 - 1.
Note: This LoRA doesn’t include any Digimon-related data, so don’t try forcing it. Some checkpoints already contain a lot of Digimon info by default—you can rely on those to generate Digimon. But combining them with this LoRA is a bit of a gamble. It might work, or it might not.
I recommend using 1344×1344 as a base resolution, like 1152×1536 or 1024×1536. The chances of getting the correct body size differences are higher when generating “2boys,” while “3boys” tends to be much harder to control. If things aren’t working, it might be an issue with prompt structure—try tweaking, adding, or removing prompts.
I won’t go into other generation parameters here since nothing else really stood out so far.
As for the training parameter overview, I won’t be including it anymore. I’ve finally found a relatively stable method and setup for creating a “2boys LoRA” from scratch, and I’m currently working on the second part of my personal notes, which will cover all the details. It should be ready in early August—if you’re interested, feel free to check it on my tensor(https://tensor.art/u/752354317775824679).
See you next time~
1. The rights to reposted models belong to original creators.
2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights. Click to Claim
