From 253cb313625027db8dd105cc80afc7025e9c1c6f Mon Sep 17 00:00:00 2001 From: Jaret Burkett Date: Sun, 22 Mar 2026 22:09:23 -0600 Subject: [PATCH] Fix issue with video and images with no audio on ltx models --- extensions_built_in/diffusion_models/ltx2/ltx2.py | 13 ++++++++----- version.py | 2 +- 2 files changed, 9 insertions(+), 6 deletions(-) diff --git a/extensions_built_in/diffusion_models/ltx2/ltx2.py b/extensions_built_in/diffusion_models/ltx2/ltx2.py index 96412f6..ea81965 100644 --- a/extensions_built_in/diffusion_models/ltx2/ltx2.py +++ b/extensions_built_in/diffusion_models/ltx2/ltx2.py @@ -909,14 +909,17 @@ class LTX2Model(BaseModel): num_channels_latents_audio = ( self.pipeline.audio_vae.config.latent_channels ) - audio_latents, audio_num_frames = self.pipeline.prepare_audio_latents( + duration_s = batch.dataset_config.num_frames / frame_rate + audio_latents_per_second = ( + self.pipeline.audio_sampling_rate / self.pipeline.audio_hop_length / float(self.pipeline.audio_vae_temporal_compression_ratio) + ) + audio_num_frames = round(duration_s * audio_latents_per_second) + audio_latents = self.pipeline.prepare_audio_latents( batch_size, num_channels_latents=num_channels_latents_audio, + audio_latent_length=audio_num_frames, num_mel_bins=num_mel_bins, - num_frames=batch.dataset_config.num_frames, - frame_rate=frame_rate, - sampling_rate=self.pipeline.audio_sampling_rate, - hop_length=self.pipeline.audio_hop_length, + noise_scale=0.0, dtype=torch.float32, device=self.transformer.device, generator=None, diff --git a/version.py b/version.py index 2a80d12..4106d5f 100644 --- a/version.py +++ b/version.py @@ -1 +1 @@ -VERSION = "0.7.26" +VERSION = "0.7.27"