The AniMax Engine is an integrated modelling system that combines several elements of diffusion-based generation (refiner, ControlNet, hires fix, adetail).The double clips (text encoders) have been retrained specifically for the model.The double text encoder layers have a cross-attention scale value of 0.6-0. 7 optimized so that text instructions do not clash with visual patterns

Double timing

The key to style preservation is when to activate the refiners:

First 50% generation: only the rough anatomical skeleton (e.g. pelvic-shoulder line)

Last 50%: ControlNet (OpenPose) and the refiner work in parallel, so details (e.g. muscle tone) do not overwrite the basic structure.

The secret is not in a magic promt, but in making body composition the primary constraint at every stage of generation. Other models often "forget" this once they start adding textures or shadows.