This is a LoRA that merges the characters from ROBOT ON THE ROAD.

You can summon the main character Mina, Robo, and the two women who appeared in the ending.

I love this anime and had been wanting to create a LoRA for it. I started by creating LoRAs for individual characters, gained experience through that process, and finally created an All-in-One LoRA. I enlisted the help of my friends for the production and testing. I’d like to take this opportunity to thank them.

Characters

Mina (Trigger: rotrmina)

The protagonist of the anime. She’s a woman traveling in a camper trailer who picked up Robo while he was hitchhiking.

Robo (Trigger: rotrrobo)

A robot who was hitchhiking. He’s a lecherous robot who hitches rides with beautiful women, takes photos of them while they’re asleep, and commits sexual pranks.

Traveler A (Trigger:rotrkita)

The woman in the camper van who appears in the anime’s ending sequence; she was Robo’s next target.

Traveler B (Trigger:rotrnisi)

A woman who was riding along in Traveler A’s camper van.

Update History

V1.0

This is the first version. For the most part, the characters can be distinguished, but occasionally Mina’s features may appear mixed in. This is most noticeable in the shape of the bangs. Also, the probability of this happening seems to increase when the women are rendered completely naked.

I imagine this is likely because there were no fully nude datasets for Travelers A and B. (Mina appears nude at certain points in the anime, so a dataset for her was available.)

You can use the separately released LoRA models for Traveler A and Traveler B in combination to adjust the intensity and so on.

I plan to investigate and fix this in the future if I have time, but for now, I’m off to raise a toast with my friends.

It was a lot of work to create this…

V1.1

We added nude and swimsuit images of Travelers A and B to the dataset and retrained the model.

As a result, it appears that these two individuals’ hairstyles and facial features are now better reflected than in the previous version, even when they are not wearing clothes.

We used 35 images per person in the dataset.

While Mina had 10 nude and swimsuit images in the dataset, Travelers A and B had none.

We adjusted their datasets, added the same number of nude images as Mina had, and retrained the model.

Even so, their features still occasionally get mixed up. Since consistency can be maintained by combining this with a standalone LoRA for each character and adjusting the weighting, you might want to try that approach depending on the image you want to generate.

I may post another update at some point.