Fix FlashVSR frame padding to match pipeline requirements

The pipeline requires num_frames % 4 == 1. Our old _pad_video_5d used a
wrong formula that produced non-conforming counts (e.g. 33 input → 35
padded → pipeline rounds to 37, wasting VRAM).

New padding uses num_frames % 8 == 1 (also satisfies % 4 == 1), which
ensures the streaming loop output exactly matches num_frames with zero
waste. Optimal input counts: 25, 33, 41, 49, 57, 65, 73, 81, 89, 97, 105.

Also removes incorrect 2-frame warmup stripping from _restore_video_sequence
— the pipeline output doesn't have warmup artifacts.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

inference.py

@@ -728,30 +728,28 @@ class FlashVSRModel:
 
     @staticmethod
     def _pad_video_5d(video):
-        """Pad [1, C, F, H, W] video: repeat last 2 frames, align for pipeline.
+        """Pad [1, C, F, H, W] video for pipeline (num_frames % 8 == 1, min 25).
 
-        Uses the reference formula: (F_padded + 2 - 5) % 8 == 0, ensuring
+        The pipeline requires num_frames % 4 == 1 and the streaming loop
-        the pipeline's streaming loop gets correct iteration counts.
+        produces 8*P+17 output frames where P = (num_frames-1)//8 - 2.
+        Using % 8 == 1 ensures output == num_frames (no wasted iterations).
+
+        Optimal input frame counts (zero padding): 25, 33, 41, 49, 57, 65, ...
         """
-        tail = video[:, :, -1:].repeat(1, 1, 2, 1, 1)
+        n = video.shape[2]
-        video = torch.cat([video, tail], dim=2)
+        target = max(n, 25)  # minimum for streaming loop (P >= 1)
-        added = 0
+        remainder = (target - 1) % 8
-        remainder = (video.shape[2] + 2 - 5) % 8
         if remainder != 0:
-            added = 8 - remainder
+            target += 8 - remainder
+        added = target - n
+        if added > 0:
             pad = video[:, :, -1:].repeat(1, 1, added, 1, 1)
             video = torch.cat([video, pad], dim=2)
         return video, added
 
     @staticmethod
     def _restore_video_sequence(result, added_frames, expected):
-        """Strip padding and warmup frames from the output."""
+        """Trim pipeline output to the expected frame count."""
-        if added_frames > 0 and result.shape[0] > added_frames:
-            result = result[:-added_frames]
-        # Strip the first 2 pipeline warmup frames
-        if result.shape[0] > 2:
-            result = result[2:]
-        # Adjust to exact expected count
         if result.shape[0] > expected:
             result = result[:expected]
         elif result.shape[0] < expected:

nodes.py

@@ -1731,14 +1731,14 @@ class FlashVSRUpscale:
                 chunks.append((prev_start, last_end))
 
         # Estimate total pipeline steps for progress bar
-        # Mirrors _pad_video_5d: add 2 tail frames, then align with (F+2-5)%8
+        # Mirrors _pad_video_5d: pad to num_frames % 8 == 1, min 25
         total_steps = 0
         for cs, ce in chunks:
-            padded_n = (ce - cs) + 2  # tail frames appended by _pad_video_5d
+            nf = max(ce - cs, 25)
-            remainder = (padded_n + 2 - 5) % 8
+            remainder = (nf - 1) % 8
             if remainder != 0:
-                padded_n += 8 - remainder
+                nf += 8 - remainder
-            total_steps += max(1, (padded_n - 1) // 8 - 2)
+            total_steps += max(1, (nf - 1) // 8 - 2)
 
         pbar = ProgressBar(total_steps)
         step_ref = [0]