The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Fix LTXV Reference Audio node (#13531)
Jukka Seppänen6fbb6b6f49cc
diff --git a/comfy_extras/nodes_lt.py b/comfy_extras/nodes_lt.pyindex d7c2e8744..19d8a387f 100644--- a/comfy_extras/nodes_lt.py+++ b/comfy_extras/nodes_lt.py@@ -1,6 +1,7 @@import nodesimport node_helpersimport torch+import torchaudioimport comfy.model_managementimport comfy.model_samplingimport comfy.samplers@@ -711,7 +712,14 @@ class LTXVReferenceAudio(io.ComfyNode):@classmethoddef execute(cls, model, positive, negative, reference_audio, audio_vae, identity_guidance_scale, start_percent, end_percent) -> io.NodeOutput:# Encode reference audio to latents and patchify- audio_latents = audio_vae.encode(reference_audio)+ sample_rate = reference_audio["sample_rate"]+ vae_sample_rate = getattr(audio_vae, "audio_sample_rate", 44100)+ if vae_sample_rate != sample_rate:+ waveform = torchaudio.functional.resample(reference_audio["waveform"], sample_rate, vae_sample_rate)+ else:+ waveform = reference_audio["waveform"]++ audio_latents = audio_vae.encode(waveform.movedim(1, -1))b, c, t, f = audio_latents.shaperef_tokens = audio_latents.permute(0, 2, 1, 3).reshape(b, t, c * f)ref_audio = {"tokens": ref_tokens}