Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Support standalone LTXV audio VAEs (#13499)

Jukka Seppäneneb2222538739
diff --git a/comfy/sd.py b/comfy/sd.py
index a4d3ee269..736fe35de 100644
--- a/comfy/sd.py
+++ b/comfy/sd.py
@@ -807,6 +807,7 @@ class VAE:
self.memory_used_encode = lambda shape, dtype: (700 * (max(1, (shape[-3] ** 0.66 * 0.11)) * shape[-2] * shape[-1]) * model_management.dtype_size(dtype))
self.memory_used_decode = lambda shape, dtype: (50 * (max(1, (shape[-3] ** 0.65 * 0.26)) * shape[-2] * shape[-1] * 32 * 32) * model_management.dtype_size(dtype))
elif "vocoder.resblocks.0.convs1.0.weight" in sd or "vocoder.vocoder.resblocks.0.convs1.0.weight" in sd: # LTX Audio
+ sd = comfy.utils.state_dict_prefix_replace(sd, {"audio_vae.": "autoencoder."})
self.first_stage_model = comfy.ldm.lightricks.vae.audio_vae.AudioVAE(metadata=metadata)
self.memory_used_encode = lambda shape, dtype: (shape[2] * 330) * model_management.dtype_size(dtype)
self.memory_used_decode = lambda shape, dtype: (shape[2] * shape[3] * 87000) * model_management.dtype_size(dtype)