Marco
793402f93f
This commit introduces several new performance-oriented custom nodes and utilities: - Added LTXV Multi Concat and LTXV Multi Concat (beta) for faster frame injection using latent inpainting logic. - Refactored LTXV index resolution logic into ltxv_utils.py. - Updated ImageListSampler to anchor edge frames properly and support output index normalization scaling via target_frames. - Added metadata-based prompt extraction image loaders (LoadImageAndExtractPrompt, FolderImageAndExtractPrompt, and FolderImageMetadataByName). - Included animation tracking and camera switcher nodes.
86 lines
3.1 KiB
Python
86 lines
3.1 KiB
Python
import torch
|
|
|
|
def resolve_frame_indices(indices, positions_str, mode, fps, total_pixel_frames, num_images):
|
|
"""Resolve frame indices from either INT list or positions string."""
|
|
# Priority 1: indices input (from ImageListSampler etc.)
|
|
if indices is not None and len(indices) > 0 and indices[0] is not None:
|
|
if mode == "frames":
|
|
return [int(i) for i in indices]
|
|
elif mode == "seconds":
|
|
return [
|
|
-1 if i < 0 else min(round(float(i) / fps * fps), total_pixel_frames - 1)
|
|
if False else int(i) # indices are already frame numbers
|
|
for i in indices
|
|
]
|
|
elif mode == "percentage":
|
|
return [
|
|
-1 if i >= total_pixel_frames - 1
|
|
else int(i)
|
|
for i in indices
|
|
]
|
|
# Default: treat indices as frame numbers directly
|
|
return [int(i) for i in indices]
|
|
|
|
# Priority 2: positions string
|
|
if positions_str and positions_str.strip():
|
|
return parse_positions(positions_str, mode, fps, total_pixel_frames)
|
|
|
|
# Fallback: distribute evenly
|
|
if num_images == 1:
|
|
return [0]
|
|
step = (total_pixel_frames - 1) / (num_images - 1)
|
|
return [round(i * step) for i in range(num_images)]
|
|
|
|
def parse_positions(positions_str, mode, fps, total_pixel_frames):
|
|
"""Parse position string into frame indices."""
|
|
raw = [s.strip() for s in positions_str.split(",") if s.strip()]
|
|
frame_indices = []
|
|
|
|
for val_str in raw:
|
|
val = float(val_str)
|
|
|
|
if mode == "frames":
|
|
frame_indices.append(int(val))
|
|
elif mode == "seconds":
|
|
if val < 0:
|
|
frame_indices.append(-1)
|
|
else:
|
|
frame_idx = round(val * fps)
|
|
frame_idx = min(frame_idx, total_pixel_frames - 1)
|
|
frame_indices.append(frame_idx)
|
|
elif mode == "percentage":
|
|
if val < 0 or val >= 100.0:
|
|
frame_indices.append(-1)
|
|
else:
|
|
frame_idx = round(val / 100.0 * (total_pixel_frames - 1))
|
|
frame_indices.append(frame_idx)
|
|
|
|
return frame_indices
|
|
|
|
def parse_strengths(strengths_str, num_images, default_strength):
|
|
"""Parse per-image strengths or fill with default."""
|
|
if not strengths_str or not strengths_str.strip():
|
|
return [default_strength] * num_images
|
|
|
|
raw = [s.strip() for s in strengths_str.split(",") if s.strip()]
|
|
parsed = [max(0.0, min(1.0, float(s))) for s in raw]
|
|
|
|
while len(parsed) < num_images:
|
|
parsed.append(default_strength)
|
|
|
|
return parsed[:num_images]
|
|
|
|
def flatten_images(images):
|
|
"""Flatten list or batch of images into a single B,H,W,C tensor."""
|
|
if isinstance(images, list):
|
|
frames = []
|
|
for item in images:
|
|
if isinstance(item, torch.Tensor):
|
|
if len(item.shape) == 4:
|
|
for j in range(item.shape[0]):
|
|
frames.append(item[j:j+1])
|
|
else:
|
|
frames.append(item.unsqueeze(0) if len(item.shape) == 3 else item)
|
|
return torch.cat(frames, dim=0) if frames else images[0]
|
|
return images
|