Update app.py
Browse files
app.py
CHANGED
|
@@ -11,7 +11,6 @@ import tempfile
|
|
| 11 |
from PIL import Image
|
| 12 |
from huggingface_hub import hf_hub_download
|
| 13 |
import shutil
|
| 14 |
-
import math # For math.round, though built-in round works for floats
|
| 15 |
|
| 16 |
from inference import (
|
| 17 |
create_ltx_video_pipeline,
|
|
@@ -89,13 +88,56 @@ if PIPELINE_CONFIG_YAML.get("spatial_upscaler_model_path"):
|
|
| 89 |
target_inference_device = "cuda"
|
| 90 |
print(f"Target inference device: {target_inference_device}")
|
| 91 |
pipeline_instance.to(target_inference_device)
|
| 92 |
-
if latent_upsampler_instance:
|
| 93 |
latent_upsampler_instance.to(target_inference_device)
|
| 94 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
@spaces.GPU
|
| 96 |
def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath,
|
| 97 |
height_ui, width_ui, mode,
|
| 98 |
-
ui_steps, duration_ui,
|
| 99 |
ui_frames_to_use,
|
| 100 |
seed_ui, randomize_seed, ui_guidance_scale, improve_texture_flag,
|
| 101 |
progress=gr.Progress(track_tqdm=True)):
|
|
@@ -104,33 +146,25 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 104 |
seed_ui = random.randint(0, 2**32 - 1)
|
| 105 |
seed_everething(int(seed_ui))
|
| 106 |
|
| 107 |
-
# Convert duration_ui (seconds) to actual_num_frames (N*8+1 format)
|
| 108 |
target_frames_ideal = duration_ui * FPS
|
| 109 |
target_frames_rounded = round(target_frames_ideal)
|
| 110 |
-
if target_frames_rounded < 1:
|
| 111 |
target_frames_rounded = 1
|
| 112 |
|
| 113 |
-
# Calculate N for N*8+1, ensuring it's rounded to the nearest integer
|
| 114 |
-
# (target_frames_rounded - 1) could be float if target_frames_rounded is float
|
| 115 |
n_val = round((float(target_frames_rounded) - 1.0) / 8.0)
|
| 116 |
actual_num_frames = int(n_val * 8 + 1)
|
| 117 |
|
| 118 |
-
# Clamp to the allowed min (9) and max (MAX_NUM_FRAMES) N*8+1 values
|
| 119 |
actual_num_frames = max(9, actual_num_frames)
|
| 120 |
actual_num_frames = min(MAX_NUM_FRAMES, actual_num_frames)
|
| 121 |
|
| 122 |
actual_height = int(height_ui)
|
| 123 |
actual_width = int(width_ui)
|
| 124 |
-
# actual_num_frames is now calculated above
|
| 125 |
|
| 126 |
height_padded = ((actual_height - 1) // 32 + 1) * 32
|
| 127 |
width_padded = ((actual_width - 1) // 32 + 1) * 32
|
| 128 |
-
# This padding ensures the model gets a frame count that is N*8+1
|
| 129 |
-
# Since actual_num_frames is already N*8+1, this should preserve it.
|
| 130 |
num_frames_padded = ((actual_num_frames - 2) // 8 + 1) * 8 + 1
|
| 131 |
if num_frames_padded != actual_num_frames:
|
| 132 |
print(f"Warning: actual_num_frames ({actual_num_frames}) and num_frames_padded ({num_frames_padded}) differ. Using num_frames_padded for pipeline.")
|
| 133 |
-
# This case should ideally not happen if actual_num_frames is correctly N*8+1 and >= 9.
|
| 134 |
|
| 135 |
padding_values = calculate_padding(actual_height, actual_width, height_padded, width_padded)
|
| 136 |
|
|
@@ -139,7 +173,7 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 139 |
"negative_prompt": negative_prompt,
|
| 140 |
"height": height_padded,
|
| 141 |
"width": width_padded,
|
| 142 |
-
"num_frames": num_frames_padded,
|
| 143 |
"frame_rate": int(FPS),
|
| 144 |
"generator": torch.Generator(device=target_inference_device).manual_seed(int(seed_ui)),
|
| 145 |
"output_type": "pt",
|
|
@@ -184,7 +218,7 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 184 |
media_path=input_video_filepath,
|
| 185 |
height=actual_height,
|
| 186 |
width=actual_width,
|
| 187 |
-
max_frames=int(ui_frames_to_use),
|
| 188 |
padding=padding_values
|
| 189 |
).to(target_inference_device)
|
| 190 |
except Exception as e:
|
|
@@ -192,15 +226,10 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 192 |
raise gr.Error(f"Could not load video: {e}")
|
| 193 |
|
| 194 |
print(f"Moving models to {target_inference_device} for inference (if not already there)...")
|
| 195 |
-
# Models are moved globally once, no need to move per call unless strategy changes.
|
| 196 |
-
# pipeline_instance.to(target_inference_device)
|
| 197 |
-
# if latent_upsampler_instance:
|
| 198 |
-
# latent_upsampler_instance.to(target_inference_device)
|
| 199 |
|
| 200 |
active_latent_upsampler = None
|
| 201 |
if improve_texture_flag and latent_upsampler_instance:
|
| 202 |
active_latent_upsampler = latent_upsampler_instance
|
| 203 |
-
#print("Models moved.")
|
| 204 |
|
| 205 |
result_images_tensor = None
|
| 206 |
if improve_texture_flag:
|
|
@@ -230,7 +259,6 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 230 |
single_pass_call_kwargs = call_kwargs.copy()
|
| 231 |
single_pass_call_kwargs["guidance_scale"] = float(ui_guidance_scale)
|
| 232 |
single_pass_call_kwargs["num_inference_steps"] = int(ui_steps)
|
| 233 |
-
# These keys might not exist if improve_texture_flag is false from the start of call_kwargs
|
| 234 |
single_pass_call_kwargs.pop("first_pass", None)
|
| 235 |
single_pass_call_kwargs.pop("second_pass", None)
|
| 236 |
single_pass_call_kwargs.pop("downscale_factor", None)
|
|
@@ -245,7 +273,6 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 245 |
slice_h_end = -pad_bottom if pad_bottom > 0 else None
|
| 246 |
slice_w_end = -pad_right if pad_right > 0 else None
|
| 247 |
|
| 248 |
-
# Crop to actual_num_frames, which is the desired output length
|
| 249 |
result_images_tensor = result_images_tensor[
|
| 250 |
:, :, :actual_num_frames, pad_top:slice_h_end, pad_left:slice_w_end
|
| 251 |
]
|
|
@@ -297,6 +324,7 @@ def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath
|
|
| 297 |
|
| 298 |
return output_video_path
|
| 299 |
|
|
|
|
| 300 |
# --- Gradio UI Definition ---
|
| 301 |
css="""
|
| 302 |
#col-container {
|
|
@@ -308,6 +336,7 @@ css="""
|
|
| 308 |
with gr.Blocks(css=css) as demo:
|
| 309 |
gr.Markdown("# LTX Video 0.9.7 Distilled")
|
| 310 |
gr.Markdown("Fast high quality video generation. [Model](https://huggingface.co/Lightricks/LTX-Video/blob/main/ltxv-2b-0.9.6-distilled-04-25.safetensors) [GitHub](https://github.com/Lightricks/LTX-Video) [Diffusers](#)")
|
|
|
|
| 311 |
with gr.Row():
|
| 312 |
with gr.Column():
|
| 313 |
with gr.Tab("image-to-video") as image_tab:
|
|
@@ -322,7 +351,7 @@ with gr.Blocks(css=css) as demo:
|
|
| 322 |
t2v_button = gr.Button("Generate Text-to-Video", variant="primary")
|
| 323 |
with gr.Tab("video-to-video") as video_tab:
|
| 324 |
image_v_hidden = gr.Textbox(label="image_v", visible=False, value=None)
|
| 325 |
-
video_v2v = gr.Video(label="Input Video", sources=["upload", "webcam"])
|
| 326 |
frames_to_use = gr.Slider(label="Frames to use from input video", minimum=9, maximum=MAX_NUM_FRAMES, value=9, step=8, info="Number of initial frames to use for conditioning/transformation. Must be N*8+1.")
|
| 327 |
v2v_prompt = gr.Textbox(label="Prompt", value="Change the style to cinematic anime", lines=3)
|
| 328 |
v2v_button = gr.Button("Generate Video-to-Video", variant="primary")
|
|
@@ -347,26 +376,90 @@ with gr.Blocks(css=css) as demo:
|
|
| 347 |
randomize_seed_input = gr.Checkbox(label="Randomize Seed", value=False)
|
| 348 |
with gr.Row():
|
| 349 |
guidance_scale_input = gr.Slider(label="Guidance Scale (CFG)", minimum=1.0, maximum=10.0, value=PIPELINE_CONFIG_YAML.get("first_pass", {}).get("guidance_scale", 1.0), step=0.1, info="Controls how much the prompt influences the output. Higher values = stronger influence.")
|
| 350 |
-
default_steps = len(PIPELINE_CONFIG_YAML.get("first_pass", {}).get("timesteps", [1]*7))
|
| 351 |
steps_input = gr.Slider(label="Inference Steps (for first pass if multi-scale)", minimum=1, maximum=30, value=default_steps, step=1, info="Number of denoising steps. More steps can improve quality but increase time. If YAML defines 'timesteps' for a pass, this UI value is ignored for that pass.")
|
| 352 |
with gr.Row():
|
| 353 |
-
height_input = gr.Slider(label="Height", value=512, step=32, minimum=
|
| 354 |
-
width_input = gr.Slider(label="Width", value=704, step=32, minimum=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 355 |
|
| 356 |
-
# ---
|
| 357 |
t2v_inputs = [t2v_prompt, negative_prompt_input, image_n_hidden, video_n_hidden,
|
| 358 |
height_input, width_input, gr.State("text-to-video"),
|
| 359 |
-
steps_input, duration_input, gr.State(0),
|
| 360 |
seed_input, randomize_seed_input, guidance_scale_input, improve_texture]
|
| 361 |
|
| 362 |
i2v_inputs = [i2v_prompt, negative_prompt_input, image_i2v, video_i_hidden,
|
| 363 |
height_input, width_input, gr.State("image-to-video"),
|
| 364 |
-
steps_input, duration_input, gr.State(0),
|
| 365 |
seed_input, randomize_seed_input, guidance_scale_input, improve_texture]
|
| 366 |
|
| 367 |
v2v_inputs = [v2v_prompt, negative_prompt_input, image_v_hidden, video_v2v,
|
| 368 |
height_input, width_input, gr.State("video-to-video"),
|
| 369 |
-
steps_input, duration_input, frames_to_use,
|
| 370 |
seed_input, randomize_seed_input, guidance_scale_input, improve_texture]
|
| 371 |
|
| 372 |
t2v_button.click(fn=generate, inputs=t2v_inputs, outputs=[output_video], api_name="text_to_video")
|
|
|
|
| 11 |
from PIL import Image
|
| 12 |
from huggingface_hub import hf_hub_download
|
| 13 |
import shutil
|
|
|
|
| 14 |
|
| 15 |
from inference import (
|
| 16 |
create_ltx_video_pipeline,
|
|
|
|
| 88 |
target_inference_device = "cuda"
|
| 89 |
print(f"Target inference device: {target_inference_device}")
|
| 90 |
pipeline_instance.to(target_inference_device)
|
| 91 |
+
if latent_upsampler_instance:
|
| 92 |
latent_upsampler_instance.to(target_inference_device)
|
| 93 |
|
| 94 |
+
|
| 95 |
+
# --- Helper function for dimension calculation ---
|
| 96 |
+
MIN_DIM_SLIDER = 256 # As defined in the sliders minimum attribute
|
| 97 |
+
TARGET_FIXED_SIDE = 512 # Desired fixed side length as per requirement
|
| 98 |
+
|
| 99 |
+
def calculate_new_dimensions(orig_w, orig_h):
|
| 100 |
+
"""
|
| 101 |
+
Calculates new dimensions for height and width sliders based on original media dimensions.
|
| 102 |
+
Ensures one side is TARGET_FIXED_SIDE, the other is scaled proportionally,
|
| 103 |
+
both are multiples of 32, and within [MIN_DIM_SLIDER, MAX_IMAGE_SIZE].
|
| 104 |
+
"""
|
| 105 |
+
if orig_w == 0 or orig_h == 0:
|
| 106 |
+
# Default to TARGET_FIXED_SIDE square if original dimensions are invalid
|
| 107 |
+
return int(TARGET_FIXED_SIDE), int(TARGET_FIXED_SIDE)
|
| 108 |
+
|
| 109 |
+
if orig_w >= orig_h: # Landscape or square
|
| 110 |
+
new_h = TARGET_FIXED_SIDE
|
| 111 |
+
aspect_ratio = orig_w / orig_h
|
| 112 |
+
new_w_ideal = new_h * aspect_ratio
|
| 113 |
+
|
| 114 |
+
# Round to nearest multiple of 32
|
| 115 |
+
new_w = round(new_w_ideal / 32) * 32
|
| 116 |
+
|
| 117 |
+
# Clamp to [MIN_DIM_SLIDER, MAX_IMAGE_SIZE]
|
| 118 |
+
new_w = max(MIN_DIM_SLIDER, min(new_w, MAX_IMAGE_SIZE))
|
| 119 |
+
# Ensure new_h is also clamped (TARGET_FIXED_SIDE should be within these bounds if configured correctly)
|
| 120 |
+
new_h = max(MIN_DIM_SLIDER, min(new_h, MAX_IMAGE_SIZE))
|
| 121 |
+
else: # Portrait
|
| 122 |
+
new_w = TARGET_FIXED_SIDE
|
| 123 |
+
aspect_ratio = orig_h / orig_w # Use H/W ratio for portrait scaling
|
| 124 |
+
new_h_ideal = new_w * aspect_ratio
|
| 125 |
+
|
| 126 |
+
# Round to nearest multiple of 32
|
| 127 |
+
new_h = round(new_h_ideal / 32) * 32
|
| 128 |
+
|
| 129 |
+
# Clamp to [MIN_DIM_SLIDER, MAX_IMAGE_SIZE]
|
| 130 |
+
new_h = max(MIN_DIM_SLIDER, min(new_h, MAX_IMAGE_SIZE))
|
| 131 |
+
# Ensure new_w is also clamped
|
| 132 |
+
new_w = max(MIN_DIM_SLIDER, min(new_w, MAX_IMAGE_SIZE))
|
| 133 |
+
|
| 134 |
+
return int(new_h), int(new_w)
|
| 135 |
+
|
| 136 |
+
|
| 137 |
@spaces.GPU
|
| 138 |
def generate(prompt, negative_prompt, input_image_filepath, input_video_filepath,
|
| 139 |
height_ui, width_ui, mode,
|
| 140 |
+
ui_steps, duration_ui,
|
| 141 |
ui_frames_to_use,
|
| 142 |
seed_ui, randomize_seed, ui_guidance_scale, improve_texture_flag,
|
| 143 |
progress=gr.Progress(track_tqdm=True)):
|
|
|
|
| 146 |
seed_ui = random.randint(0, 2**32 - 1)
|
| 147 |
seed_everething(int(seed_ui))
|
| 148 |
|
|
|
|
| 149 |
target_frames_ideal = duration_ui * FPS
|
| 150 |
target_frames_rounded = round(target_frames_ideal)
|
| 151 |
+
if target_frames_rounded < 1:
|
| 152 |
target_frames_rounded = 1
|
| 153 |
|
|
|
|
|
|
|
| 154 |
n_val = round((float(target_frames_rounded) - 1.0) / 8.0)
|
| 155 |
actual_num_frames = int(n_val * 8 + 1)
|
| 156 |
|
|
|
|
| 157 |
actual_num_frames = max(9, actual_num_frames)
|
| 158 |
actual_num_frames = min(MAX_NUM_FRAMES, actual_num_frames)
|
| 159 |
|
| 160 |
actual_height = int(height_ui)
|
| 161 |
actual_width = int(width_ui)
|
|
|
|
| 162 |
|
| 163 |
height_padded = ((actual_height - 1) // 32 + 1) * 32
|
| 164 |
width_padded = ((actual_width - 1) // 32 + 1) * 32
|
|
|
|
|
|
|
| 165 |
num_frames_padded = ((actual_num_frames - 2) // 8 + 1) * 8 + 1
|
| 166 |
if num_frames_padded != actual_num_frames:
|
| 167 |
print(f"Warning: actual_num_frames ({actual_num_frames}) and num_frames_padded ({num_frames_padded}) differ. Using num_frames_padded for pipeline.")
|
|
|
|
| 168 |
|
| 169 |
padding_values = calculate_padding(actual_height, actual_width, height_padded, width_padded)
|
| 170 |
|
|
|
|
| 173 |
"negative_prompt": negative_prompt,
|
| 174 |
"height": height_padded,
|
| 175 |
"width": width_padded,
|
| 176 |
+
"num_frames": num_frames_padded,
|
| 177 |
"frame_rate": int(FPS),
|
| 178 |
"generator": torch.Generator(device=target_inference_device).manual_seed(int(seed_ui)),
|
| 179 |
"output_type": "pt",
|
|
|
|
| 218 |
media_path=input_video_filepath,
|
| 219 |
height=actual_height,
|
| 220 |
width=actual_width,
|
| 221 |
+
max_frames=int(ui_frames_to_use),
|
| 222 |
padding=padding_values
|
| 223 |
).to(target_inference_device)
|
| 224 |
except Exception as e:
|
|
|
|
| 226 |
raise gr.Error(f"Could not load video: {e}")
|
| 227 |
|
| 228 |
print(f"Moving models to {target_inference_device} for inference (if not already there)...")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 229 |
|
| 230 |
active_latent_upsampler = None
|
| 231 |
if improve_texture_flag and latent_upsampler_instance:
|
| 232 |
active_latent_upsampler = latent_upsampler_instance
|
|
|
|
| 233 |
|
| 234 |
result_images_tensor = None
|
| 235 |
if improve_texture_flag:
|
|
|
|
| 259 |
single_pass_call_kwargs = call_kwargs.copy()
|
| 260 |
single_pass_call_kwargs["guidance_scale"] = float(ui_guidance_scale)
|
| 261 |
single_pass_call_kwargs["num_inference_steps"] = int(ui_steps)
|
|
|
|
| 262 |
single_pass_call_kwargs.pop("first_pass", None)
|
| 263 |
single_pass_call_kwargs.pop("second_pass", None)
|
| 264 |
single_pass_call_kwargs.pop("downscale_factor", None)
|
|
|
|
| 273 |
slice_h_end = -pad_bottom if pad_bottom > 0 else None
|
| 274 |
slice_w_end = -pad_right if pad_right > 0 else None
|
| 275 |
|
|
|
|
| 276 |
result_images_tensor = result_images_tensor[
|
| 277 |
:, :, :actual_num_frames, pad_top:slice_h_end, pad_left:slice_w_end
|
| 278 |
]
|
|
|
|
| 324 |
|
| 325 |
return output_video_path
|
| 326 |
|
| 327 |
+
|
| 328 |
# --- Gradio UI Definition ---
|
| 329 |
css="""
|
| 330 |
#col-container {
|
|
|
|
| 336 |
with gr.Blocks(css=css) as demo:
|
| 337 |
gr.Markdown("# LTX Video 0.9.7 Distilled")
|
| 338 |
gr.Markdown("Fast high quality video generation. [Model](https://huggingface.co/Lightricks/LTX-Video/blob/main/ltxv-2b-0.9.6-distilled-04-25.safetensors) [GitHub](https://github.com/Lightricks/LTX-Video) [Diffusers](#)")
|
| 339 |
+
|
| 340 |
with gr.Row():
|
| 341 |
with gr.Column():
|
| 342 |
with gr.Tab("image-to-video") as image_tab:
|
|
|
|
| 351 |
t2v_button = gr.Button("Generate Text-to-Video", variant="primary")
|
| 352 |
with gr.Tab("video-to-video") as video_tab:
|
| 353 |
image_v_hidden = gr.Textbox(label="image_v", visible=False, value=None)
|
| 354 |
+
video_v2v = gr.Video(label="Input Video", sources=["upload", "webcam"]) # type defaults to filepath
|
| 355 |
frames_to_use = gr.Slider(label="Frames to use from input video", minimum=9, maximum=MAX_NUM_FRAMES, value=9, step=8, info="Number of initial frames to use for conditioning/transformation. Must be N*8+1.")
|
| 356 |
v2v_prompt = gr.Textbox(label="Prompt", value="Change the style to cinematic anime", lines=3)
|
| 357 |
v2v_button = gr.Button("Generate Video-to-Video", variant="primary")
|
|
|
|
| 376 |
randomize_seed_input = gr.Checkbox(label="Randomize Seed", value=False)
|
| 377 |
with gr.Row():
|
| 378 |
guidance_scale_input = gr.Slider(label="Guidance Scale (CFG)", minimum=1.0, maximum=10.0, value=PIPELINE_CONFIG_YAML.get("first_pass", {}).get("guidance_scale", 1.0), step=0.1, info="Controls how much the prompt influences the output. Higher values = stronger influence.")
|
| 379 |
+
default_steps = len(PIPELINE_CONFIG_YAML.get("first_pass", {}).get("timesteps", [1]*7))
|
| 380 |
steps_input = gr.Slider(label="Inference Steps (for first pass if multi-scale)", minimum=1, maximum=30, value=default_steps, step=1, info="Number of denoising steps. More steps can improve quality but increase time. If YAML defines 'timesteps' for a pass, this UI value is ignored for that pass.")
|
| 381 |
with gr.Row():
|
| 382 |
+
height_input = gr.Slider(label="Height", value=512, step=32, minimum=MIN_DIM_SLIDER, maximum=MAX_IMAGE_SIZE, info="Must be divisible by 32.")
|
| 383 |
+
width_input = gr.Slider(label="Width", value=704, step=32, minimum=MIN_DIM_SLIDER, maximum=MAX_IMAGE_SIZE, info="Must be divisible by 32.")
|
| 384 |
+
|
| 385 |
+
|
| 386 |
+
# --- Event handlers for updating dimensions on upload ---
|
| 387 |
+
def handle_image_upload_for_dims(image_filepath, current_h, current_w):
|
| 388 |
+
if not image_filepath: # Image cleared or no image initially
|
| 389 |
+
# Keep current slider values if image is cleared or no input
|
| 390 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 391 |
+
try:
|
| 392 |
+
img = Image.open(image_filepath)
|
| 393 |
+
orig_w, orig_h = img.size
|
| 394 |
+
new_h, new_w = calculate_new_dimensions(orig_w, orig_h)
|
| 395 |
+
return gr.update(value=new_h), gr.update(value=new_w)
|
| 396 |
+
except Exception as e:
|
| 397 |
+
print(f"Error processing image for dimension update: {e}")
|
| 398 |
+
# Keep current slider values on error
|
| 399 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 400 |
+
|
| 401 |
+
def handle_video_upload_for_dims(video_filepath, current_h, current_w):
|
| 402 |
+
if not video_filepath: # Video cleared or no video initially
|
| 403 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 404 |
+
try:
|
| 405 |
+
# Ensure video_filepath is a string for os.path.exists and imageio
|
| 406 |
+
video_filepath_str = str(video_filepath)
|
| 407 |
+
if not os.path.exists(video_filepath_str):
|
| 408 |
+
print(f"Video file path does not exist for dimension update: {video_filepath_str}")
|
| 409 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 410 |
+
|
| 411 |
+
orig_w, orig_h = -1, -1
|
| 412 |
+
with imageio.get_reader(video_filepath_str) as reader:
|
| 413 |
+
meta = reader.get_meta_data()
|
| 414 |
+
if 'size' in meta:
|
| 415 |
+
orig_w, orig_h = meta['size']
|
| 416 |
+
else:
|
| 417 |
+
# Fallback: read first frame if 'size' not in metadata
|
| 418 |
+
try:
|
| 419 |
+
first_frame = reader.get_data(0)
|
| 420 |
+
# Shape is (h, w, c) for frames
|
| 421 |
+
orig_h, orig_w = first_frame.shape[0], first_frame.shape[1]
|
| 422 |
+
except Exception as e_frame:
|
| 423 |
+
print(f"Could not get video size from metadata or first frame: {e_frame}")
|
| 424 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 425 |
+
|
| 426 |
+
if orig_w == -1 or orig_h == -1: # If dimensions couldn't be determined
|
| 427 |
+
print(f"Could not determine dimensions for video: {video_filepath_str}")
|
| 428 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 429 |
+
|
| 430 |
+
new_h, new_w = calculate_new_dimensions(orig_w, orig_h)
|
| 431 |
+
return gr.update(value=new_h), gr.update(value=new_w)
|
| 432 |
+
except Exception as e:
|
| 433 |
+
# Log type of video_filepath for debugging if it's not a path-like string
|
| 434 |
+
print(f"Error processing video for dimension update: {e} (Path: {video_filepath}, Type: {type(video_filepath)})")
|
| 435 |
+
return gr.update(value=current_h), gr.update(value=current_w)
|
| 436 |
+
|
| 437 |
+
# Attach upload handlers
|
| 438 |
+
image_i2v.upload(
|
| 439 |
+
fn=handle_image_upload_for_dims,
|
| 440 |
+
inputs=[image_i2v, height_input, width_input],
|
| 441 |
+
outputs=[height_input, width_input]
|
| 442 |
+
)
|
| 443 |
+
video_v2v.upload(
|
| 444 |
+
fn=handle_video_upload_for_dims,
|
| 445 |
+
inputs=[video_v2v, height_input, width_input],
|
| 446 |
+
outputs=[height_input, width_input]
|
| 447 |
+
)
|
| 448 |
|
| 449 |
+
# --- INPUT LISTS (remain the same structurally) ---
|
| 450 |
t2v_inputs = [t2v_prompt, negative_prompt_input, image_n_hidden, video_n_hidden,
|
| 451 |
height_input, width_input, gr.State("text-to-video"),
|
| 452 |
+
steps_input, duration_input, gr.State(0),
|
| 453 |
seed_input, randomize_seed_input, guidance_scale_input, improve_texture]
|
| 454 |
|
| 455 |
i2v_inputs = [i2v_prompt, negative_prompt_input, image_i2v, video_i_hidden,
|
| 456 |
height_input, width_input, gr.State("image-to-video"),
|
| 457 |
+
steps_input, duration_input, gr.State(0),
|
| 458 |
seed_input, randomize_seed_input, guidance_scale_input, improve_texture]
|
| 459 |
|
| 460 |
v2v_inputs = [v2v_prompt, negative_prompt_input, image_v_hidden, video_v2v,
|
| 461 |
height_input, width_input, gr.State("video-to-video"),
|
| 462 |
+
steps_input, duration_input, frames_to_use,
|
| 463 |
seed_input, randomize_seed_input, guidance_scale_input, improve_texture]
|
| 464 |
|
| 465 |
t2v_button.click(fn=generate, inputs=t2v_inputs, outputs=[output_video], api_name="text_to_video")
|