Refactor video generation and model loading processes to utilize from_pretrained methods for VideoEncoder and VideoDecoder. Update denoising functions to include a cfg_rescale parameter for improved artifact reduction. Ensure consistent dtype handling across audio and video processing, enhancing precision and aligning with PyTorch behavior.

mlx_video/__init__.py

@@ -10,24 +10,16 @@ from mlx_video.convert import (
 
 # Audio VAE components
 from mlx_video.models.ltx.audio_vae import (
-    AudioEncoder,
     AudioDecoder,
     Vocoder,
-    AudioProcessor,
     decode_audio,
-)
-
-# Patchifiers
-from mlx_video.components.patchifiers import (
-    VideoLatentPatchifier,
     AudioPatchifier,
-    VideoLatentShape,
     AudioLatentShape,
+    PerChannelStatistics,
 )
 
 # Conditioning
 from mlx_video.conditioning import (
-    VideoConditionByKeyframeIndex,
     VideoConditionByLatentIndex,
 )
 
@@ -43,17 +35,12 @@ __all__ = [
     "sanitize_audio_vae_weights",
     "sanitize_vocoder_weights",
     # Audio VAE
-    "AudioEncoder",
     "AudioDecoder",
     "Vocoder",
-    "AudioProcessor",
     "decode_audio",
-    # Patchifiers
-    "VideoLatentPatchifier",
     "AudioPatchifier",
-    "VideoLatentShape",
     "AudioLatentShape",
+    "PerChannelStatistics",
     # Conditioning
-    "VideoConditionByKeyframeIndex",
     "VideoConditionByLatentIndex",
 ]