revise ui from Lightricks/ltx-video-distilled (#3)

- revise ui from Lightricks/ltx-video-distilled (0dcc0bdac16da93c8d5fe44594e010baab8d5909)
- Delete pipeline_ltx_condition.py (433ae54ab7775b637c6b7a5ea236d31ee268c14d)
- ui changes part 1 (84bff9fe4140bd320824f0edd1d4924b04219022)

app.py

@@ -1,8 +1,6 @@
 import gradio as gr
 import spaces 
 import torch
-# from pipeline_ltx_condition import LTXVideoCondition, LTXConditionPipeline
-# from diffusers import LTXLatentUpsamplePipeline
 from diffusers import LTXConditionPipeline, LTXLatentUpsamplePipeline
 from diffusers.pipelines.ltx.pipeline_ltx_condition import LTXVideoCondition
 from diffusers.utils import export_to_video, load_video, load_image
@@ -26,8 +24,28 @@ def round_to_nearest_resolution_acceptable_by_vae(height, width):
     width = width - (width % pipe.vae_spatial_compression_ratio)
     print("after rounding",height, width)
     return height, width
+
+def get_duration(prompt,
+             negative_prompt,
+             image,
+             video,
+             height,
+             width,
+             mode,
+             steps,
+             num_frames,
+             frames_to_use,
+             seed,
+             randomize_seed,
+             guidance_scale,
+             duration_input,
+             improve_texture, progress):
+    if duration_input > 7:
+        return 75
+    else:
+        return 60
     
-@spaces.GPU
+@spaces.GPU(duration=get_duration)
 def generate(prompt,
              negative_prompt,
              image,
@@ -47,12 +65,18 @@ def generate(prompt,
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
 
+    # calculate number of frames based on the duration input in seconds
     target_frames_ideal = duration_input * FPS
-    num_frames = round(target_frames_ideal)
-    if num_frames < 1: 
-        num_frames = 1
+    target_frames_rounded = round(target_frames_ideal)
+    if target_frames_rounded < 1: 
+        target_frames_rounded = 1
+
+    n_val = round((float(target_frames_rounded) - 1.0) / 8.0)
+    actual_num_frames = int(n_val * 8 + 1)
+
+    actual_num_frames = max(9, actual_num_frames)
+    num_frames = min(MAX_NUM_FRAMES, actual_num_frames)
 
-    print(mode)
     if mode == "video-to-video" and (video is not None):
         video = load_video(video)[:frames_to_use]
         condition = True
@@ -161,25 +185,28 @@ function refresh() {
 with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
 
   gr.Markdown("# LTX Video 0.9.7 Distilled")
+  gr.Markdown("Fast high quality video generation. [Model](https://huggingface.co/Lightricks/LTX-Video/blob/main/ltxv-13b-0.9.7-distilled.safetensors) [GitHub](https://github.com/Lightricks/LTX-Video) [Diffusers](#)")
+    
   with gr.Row():
-    with gr.Column():
-      with gr.Group():
-        with gr.Tab("text-to-video") as text_tab:
-          image_n = gr.Image(label="", visible=False)
-          t2v_prompt = gr.Textbox(label="prompt")
-          t2v_button = gr.Button()
-        with gr.Tab("image-to-video") as image_tab:
-          image = gr.Image(label="input image", type="filepath")
-          i2v_prompt = gr.Textbox(label="prompt")
-          i2v_button = gr.Button()
-        with gr.Tab("video-to-video") as video_tab:
-          video = gr.Video(label="input video")
-          frames_to_use = gr.Number(label="num frames to use",info="first # of frames to use from the input video", value=1)
-          v2v_prompt = gr.Textbox(label="prompt")
-          v2v_button = gr.Button()
-        
-        improve_texture = gr.Checkbox(label="improve texture", value=False, info="slows down generation")
-        duration_input = gr.Slider(
+        with gr.Column():
+            with gr.Tab("image-to-video") as image_tab:
+                video_i_hidden = gr.Textbox(label="video_i", visible=False, value=None)
+                image_i2v = gr.Image(label="Input Image", type="filepath", sources=["upload", "webcam", "clipboard"])
+                i2v_prompt = gr.Textbox(label="Prompt", value="The creature from the image starts to move", lines=3)
+                i2v_button = gr.Button("Generate Image-to-Video", variant="primary")
+            with gr.Tab("text-to-video") as text_tab:
+                image_n_hidden = gr.Textbox(label="image_n", visible=False, value=None)
+                video_n_hidden = gr.Textbox(label="video_n", visible=False, value=None)
+                t2v_prompt = gr.Textbox(label="Prompt", value="A majestic dragon flying over a medieval castle", lines=3)
+                t2v_button = gr.Button("Generate Text-to-Video", variant="primary")
+            with gr.Tab("video-to-video") as video_tab:
+                image_v_hidden = gr.Textbox(label="image_v", visible=False, value=None)
+                video_v2v = gr.Video(label="Input Video")
+                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.")
+                v2v_prompt = gr.Textbox(label="Prompt", value="Change the style to cinematic anime", lines=3)
+                v2v_button = gr.Button("Generate Video-to-Video", variant="primary")
+
+            duration_input = gr.Slider(
                 label="Video Duration (seconds)", 
                 minimum=0.3, 
                 maximum=8.5, 
@@ -187,9 +214,11 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
                 step=0.1, 
                 info=f"Target video duration (0.3s to 8.5s)"
             )
+            improve_texture = gr.Checkbox(label="Improve Texture (multi-scale)", value=True, info="Uses a two-pass generation for better quality, but is slower. Recommended for final output.")
 
-    with gr.Column():
-      output = gr.Video(interactive=False)
+        with gr.Column():
+            output_video = gr.Video(label="Generated Video", interactive=False)
+            gr.DeepLinkButton()
       
 
   with gr.Accordion("Advanced settings", open=False):
@@ -200,7 +229,7 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
      with gr.Row():
       guidance_scale= gr.Slider(label="guidance scale", minimum=0, maximum=10, value=1, step=1)
       steps = gr.Slider(label="Steps", minimum=1, maximum=30, value=8, step=1)
-      num_frames = gr.Slider(label="# frames", minimum=1, maximum=161, value=96, step=1)
+      num_frames = gr.Slider(label="# frames", minimum=1, maximum=161, value=96, step=1, visible=False)
      with gr.Row():
        height = gr.Slider(label="height", value=512, step=1, maximum=2048)
        width = gr.Slider(label="width", value=704, step=1, maximum=2048)
@@ -208,8 +237,8 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
   t2v_button.click(fn=generate, 
              inputs=[t2v_prompt,
              negative_prompt,
-             image,
-             video,
+             image_n_hidden,
+             video_n_hidden,
              height,
              width,
              gr.State("text-to-video"),
@@ -217,14 +246,14 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
              num_frames,
              frames_to_use,
              seed,
-             randomize_seed,guidance_scale, improve_texture], 
+             randomize_seed,guidance_scale, duration_input, improve_texture], 
                    outputs=[output])
 
   i2v_button.click(fn=generate, 
              inputs=[i2v_prompt,
              negative_prompt,
-             image,
-             video,
+             image_i2v,
+             video_i_hidden,
              height,
              width,
              gr.State("image-to-video"),
@@ -232,14 +261,14 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
              num_frames,
              frames_to_use,
              seed,
-             randomize_seed,guidance_scale, improve_texture], 
+             randomize_seed,guidance_scale, duration_input, improve_texture], 
                    outputs=[output])
     
   v2v_button.click(fn=generate, 
              inputs=[v2v_prompt,
              negative_prompt,
-             image,
-             video,
+             image_v_hidden,
+             video_v2v,
              height,
              width,
              gr.State("video-to-video"),
@@ -247,7 +276,7 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
              num_frames,
              frames_to_use,
              seed,
-             randomize_seed,guidance_scale, improve_texture], 
+             randomize_seed,guidance_scale, duration_input, improve_texture], 
                    outputs=[output])