fix(processing): preserve pipeline audio across process_decode

process_decode strips video pipeline output to a flat list of frames at the
PIL early-return (processing_diffusers.py:461-465), so any output.audio is
lost before processing.process_images returns. video pipelines that produce
synchronized audio (LTX-2 audio-capable models) were getting silent mp4s on
the non-latent path.

stash output.audio on p.audio_capture before process_decode runs and let
processing read it back as a fallback when samples is a flat list.

ltx_process non-latent branch strips the (B, 2, N) batch dim with [0] so
write_audio's .T+contiguous() path produces interleaved bytes for AAC s16.

modules/ltx/ltx_process.py

@@ -360,8 +360,11 @@ def run_ltx(task_id,
                         yield from abort('Video: process_images returned no frames', ok=True, p=p)
                         return
                     pixels = processed.images
-                    if getattr(processed, 'audio', None) is not None:
-                        audio = processed.audio
+                    raw_audio = getattr(processed, 'audio', None)
+                    if raw_audio is not None:
+                        # Strip batch dim from (B, 2, N); write_audio expects (2, N) for the
+                        # transpose-to-interleaved path used by AAC s16.
+                        audio = raw_audio[0].float().cpu() if raw_audio.ndim == 3 else raw_audio.float().cpu()
                     latents = None
             except AssertionError as e:
                 yield from abort(e, ok=True, p=p)

modules/processing.py

@@ -528,7 +528,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
                         output_images.append(batch_image)
                         infotexts.append(batch_infotext)
 
-            audio = getattr(samples, 'audio', None)
+            audio = getattr(samples, 'audio', None) or getattr(p, 'audio_capture', None)
 
             if shared.cmd_opts.lowvram:
                 devices.torch_gc(force=True, reason='lowvram')

modules/processing_diffusers.py

@@ -628,6 +628,10 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
     timer.process.add('lora', lora_common.timer.total)
     lora_common.timer.clear(complete=True)
 
+    # process_decode flattens video output to a frame list and drops the audio attribute;
+    # stash it on `p` so video pipelines can recover it after process_images returns.
+    if output is not None and getattr(output, 'audio', None) is not None:
+        p.audio_capture = output.audio
     results = process_decode(p, output)
     timer.process.record('decode')