Files
olivv-cs-ComfyUI-FunPack/tests
digital-garbage 6e28f56c92 feat: REINS-style representation steering for H3 (experimental)
Reaches inside the forward pass instead of around it: captures the
video rows of one block's hidden state per generation, and once a
refinement key has 3+ liked and 3+ disliked runs, adds their mean
difference back into that block's output on every later run. No
gradient, no MLP -- a mean-difference of vectors is the whole model.

Masked to video rows only via the block's own mod_segments tags,
never audio or text -- the one documented failure of this class of
intervention (project_self_consistency_failed.md) happened on a
different model (LTXAV) via joint audio-video attention.

Off by default, H3-only, unvalidated.
2026-09-03 17:50:37 +03:00
..