VRM-adetail-flattener_v1

An anime flattener lora attempt using V-Roid Studio Mobile.

All training images captured and edited by me. Utilizing open copyright for non-commercial use vrms + some vrms of my own.

• Use of classroom day virtual studio.
• use of classroom night virtual studio.
• use of penthouse day studio.
• use of penthouse night studio.
• use of virtual tv set studio.
• use of tv set w/ Japanese text art.
• renders kept clean of post vfx.
• style lora designed to take hyper anime, reduce to anime flat pre application of vfx overlays.

---

Model trained on Haruka v2, will create DiTv2.0 version using same set.

style lora is experimental. Expect ranged settings. Rec settings window:

• lora weight = -1.2 to 1.2
• lora ideal weight = -0.8 to 0.8
• steps = 20 to 50
• cfg = 4.5 to 12.0
• use with face fix
• rec hand fix aid to this
• sdxl vers = illustrious fork base models
• rec as img2img pass or at least control pose + depth
• samplers rec = eulers, ddim or simple base (not kerras/ dpm sde)
• may need to play around with hires as post or pre-step.

---

further notes may follow later.