Two of 3 Mattie Love lora training testing pics, another good one with a pet i know @Tigapup likes. in these last 2 pics, they are the same prompt and same seed, only difference is i was cycling through the lora's from 100 steps to 1600 steps. i think this one was step 600, and the other is 1600 steps.
OBP used to get a basic prompt idea, then fed through Qwen2.5-7B-instruct-1M-Q8 using a specialized prompt to enhance the details, and try to get a full body shot, as well as make sure the character gets placed in the prompt. this is the result: stylized by Hideyuki Kikuchi, full body view of Mattielove ...
Workflow
comfyui used to gen the source, then used comfyui again to do a tiled upscale with FLUX Dev.