Update app.py

app.py

@@ -66,12 +66,12 @@ model = load_model()
 # Text-to-video generation function
 @spaces.GPU(duration=120)
 def generate_video(prompt, image=None, duration=5, guidance_scale=9, video_guidance_scale=5, progress=gr.Progress(track_tqdm=True)):
- multiplier = 0.8 if is_canonical else 2.4
- temp = int(duration * 0.8) # Convert seconds to temp value (assuming 24 FPS)
+ multiplier = 1.2 if is_canonical else 3.0
+ temp = int(duration * multiplier) + 1
  torch_dtype = torch.bfloat16 if MODEL_DTYPE == "bf16" else torch.float32
  if(image):
- cropped_image = center_crop(image, 1280, 720)
- resized_image = cropped_image.resize((1280, 720))
+ cropped_image = center_crop(image, 1280, 768)
+ resized_image = cropped_image.resize((1280, 768))
  with torch.no_grad(), torch.cuda.amp.autocast(enabled=True, dtype=torch_dtype):
  frames = model.generate_i2v(
  prompt=prompt,
@@ -113,12 +113,12 @@ with gr.Blocks() as demo:
  i2v_image = gr.Image(type="pil", label="Input Image")
  t2v_prompt = gr.Textbox(label="Prompt")
  with gr.Accordion("Advanced settings", open=False):
- t2v_duration = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Duration (seconds)", visible=not is_canonical)
+ t2v_duration = gr.Slider(minimum=1, maximum=3 if is_canonical else 10, value=3 if is_canonical else 5, step=1, label="Duration (seconds)", visible=not is_canonical)
  t2v_guidance_scale = gr.Slider(minimum=1, maximum=15, value=9, step=0.1, label="Guidance Scale")
  t2v_video_guidance_scale = gr.Slider(minimum=1, maximum=15, value=5, step=0.1, label="Video Guidance Scale")
  t2v_generate_btn = gr.Button("Generate Video")
  with gr.Column():
- t2v_output = gr.Video(label="Generated Video")
+ t2v_output = gr.Video(label=f"Generated Video at {'8fps' if is_caonical else '24fps'}")
  gr.HTML("""
  <div style="display: flex; flex-direction: column;justify-content: center; align-items: center; text-align: center;">
  <p style="display: flex;gap: 6px;">