v1.0
Trained on NoobAI 0.5 with 597 images
Best Results with ObsessionIllustrious checkpoint
You can use it with the character LoRa of your liking just make sure to put the weight to 0.8 to avoid artifacts and remove from the prompt: "robot" "ninja helmet"
Some character LoRas may bug out mainly because they were trained on one outfit and it can try and blend with the ranger one either way feel free to generate images so i can try and improve on the next version or future models.
Prompts:
Blue ninja: ninjablue, robot, solo, ninjablue helmet, ninjablue bodysuit, gold shoulder pads, gold belt, black pants, blue gloves, blue boots, red spear on back
Red ninja: ninjared, robot, solo, ninjared helmet, ninjared bodysuit, black belt, white gloves, red sword in back, red boots
Cyan ninja: ninjacyan, robot, solo, ninjacyan helmet, ninjacyan bodysuit, breasts, cyan skirt, black belt, white gloves, red sword in back, cyan boots
Yellow ninja: ninjayellow, robot, solo, ninjayellow helmet, ninjayellow bodysuit, black belt, white gloves, red sword in back, yellow boots
Crimson ninja: ninjacrimson, robot, solo, ninjacrimson helmet, ninjacrimson bodysuit, gold shoulder pads, gold belt, black pants, crimson gloves, crimson boots, red spear on back
Green ninja: ninjagreen, robot, solo, ninjagreen helmet, ninjagreen bodysuit, ninjagreen belt, gold vest, white gloves, green boots
Alternate Green ninja: altgreen, robot, solo, altgreen helmet, altgreen bodysuit, altgreen belt, white gloves, green boots
Tip:
you can remove the swords and spears of the back just make sure to put them in the negative prompt
Extra prompts:
Red sword: Holding red sword , red sword
Green sword: Holding green sword, green sword
Red spear: Holding red spear, red spear
Description
FAQ
Comments (3)
Not sure if its my settings but I think when this lora is on high power (~1), it tends to look a bit squished perspective wise. maybe later there should be an update to fix resolution?
mmm i can try but the show was on that sd quality so it has this squished look sometimes, maybe a thing that happened for adapting sentai footage that was on a lower resolution, i would recommend 0.8 cuz 1 gets to similar to the images in the dataset
ahh right that could be it, maybe Square generation might be better













