Intro

There's been an influx of character portraits and pixel styles recently and it made me think if something could be done with pictures of one of my favorite games, Front Mission 3. But all I found were some small non pixel art 64x80 portraits that I will train anyways to see what happens in the future, but don't give me too much hope. Feeling down I looked up the artist from those characters, that were very unique for a JRPG and that's when I found Akihiro Yamada.

His art is traditional, mostly watercolor, so a big part of the dataset is scans of pictures of paintings, so expect artifacts here and there that I will work out for v2. For this same reason and to appreciate the details of the watercolor on the generations, hires is a must.

Ads

I have a kofi if you want to support my work, from now on I'll be doing some early access to WIP LoRAs but at the end of the day, everything I do will be always free. I also take commissions, so if you are interested on a training, check that out or message me through discord. I keep a Pixiv where I upload pics from testing and tease the things I'm working on!

Tags

The LoRA comes in two flavors:

1. SD Training with Clip Skip 1, AKA for Generalist models like Deliberate and DreamShaper.

2. NAI Training with Clip Skip 2, AKA Anime models.

Why? Because it doesn't follow the modern anime aesthetic and generalist models are better at watercolor, plus, they are probably the strong ones for the second part of the contest.

The main tag is akihiro yamada which can be paired with watercolor \(medium\) and flat color depending on what result you want. There's examples and comparisons on the gallery. On anime is also good to use nose and lips so the face is not so flat. Also, thick lines helps getting better outlines. The model is capable of NSFW but to adhere to the contest rules I removed those images from the post.