This is my latest lora for Moss Image about the Splatoon character "The Mechanic," from the latest Nintendo Switch 2 game. The trigger word is "spl4t". Then just add your verbose text behind it. I usually use the prompt upsampler at the beginning of my workflow.

The dataset is around 200 images and I was careful to craft it to avoid "bleed" and "twinning." Over 6500 steps and 10 epochs to make this lora.

Sample prompt: spl4t, a girl, full-body shot at eye level in a boxing ring, standing in a fighting stance with her left leg forward and right leg back, her right arm extended punching forward while her left arm is raised defensively, mouth open in a shout with teeth showing and brows angled downward, looking at the opponent. She wears a white sleeveless tank top, red sweatpants with a white drawstring, red boxing gloves on both hands, and a teal headpiece with orange tips and a brown bun. Behind her, a referee in a grey shirt and black trousers stands in the center background watching the fight. To the left, a person holding a camera with a flash is visible. On the right, another person holds a camera near the ropes. In the foreground, a large grey creature with a bulbous head, small eyes, sharp teeth, and blue boxing gloves faces spl4t, its body covered in sweat and wearing shiny silver shorts with a blue stripe. The ring floor has colorful paint splatters and debris. Bright stadium lights shine from above, creating lens flares and illuminating water droplets in the air. The background shows a blurred crowd behind the ropes. 3D render, shallow depth of field, motion blur on the punch, cool tones with warm highlights.

Minimax-H3 Image version: same training data and steps, same trigger word, use 8 MP up to 20MP. I find that 16MP is the sweet spot!