Update README.md

README.md

@@ -9,7 +9,8 @@ tags:
 
 # MiniMax-H3 RefMod
 
-Portable reference-conditioning files for MiniMax-H3 `ref2va`, as [Modular Diffusers](https://huggingface.co/docs/diffusers/modular_diffusers/overview) blocks. This is the diffusers-modular equivalent of the ComfyUI "RefMod" trick: encode a reference once, save the result to a small `.safetensors`, and inject it straight back on later requests instead of re-encoding the media.
+Portable reference-conditioning files for MiniMax-H3 `ref2va`, as [Modular Diffusers](https://huggingface.co/docs/diffusers/modular_diffusers/overview) blocks. 
+This is the diffusers-modular equivalent of the ComfyUI "RefMod": encode a reference once, save the result to a small `.safetensors`, and inject it straight back on later requests instead of re-encoding the media.
 
 ## What it is
 
@@ -103,11 +104,4 @@ video = pipe(
 `safetensors` with:
 
 - tensors `video.0, video.1, ...` (one per image/video reference) and `audio.0, audio.1, ...` (one per soundtrack), in packed order.
-- a JSON metadata header: `format`, `version`, `num_video`, `num_audio`, `video_shapes`, `audio_shapes`, and `reference_kinds`.
-
-## Verification
-
-- **Round-trip**: a save then load returns `condition_latents` / `audio_condition_latents` that are bitwise-equal to the live encode (`max|Δ| = 0`).
-- **Drop-in**: replacing `MiniMaxH3Ref2VAReferenceEncoderStep` with `MiniMaxH3LoadRefModStep` in a full `ref2va` graph produces bitwise-identical video and audio, because the reload is exact and the reference encode never draws from the request generator.
-
-Both are verified on `hf-internal-testing/tiny-minimax-h3-modular-pipe` (shapes and numerics only; it carries random weights and says nothing about generation quality).
+- a JSON metadata header: `format`, `version`, `num_video`, `num_audio`, `video_shapes`, `audio_shapes`, and `reference_kinds`.