update app

app.py

@@ -1,4 +1,3 @@
-
 import spaces
 import torch
 from diffusers.pipelines.wan.pipeline_wan_i2v import WanImageToVideoPipeline
@@ -23,8 +22,6 @@ from torchao.quantization import Int8WeightOnlyConfig
 
 import aoti
 
-
-# ─── Model & dimension constants ───────────────────────────────────────────────
 MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
 
 MAX_DIM = 832
@@ -43,7 +40,6 @@ MAX_DURATION = round(MAX_FRAMES_MODEL / FIXED_FPS, 1)
 
 device = "cuda"
 
-# ─── Load Wan 2.2 I2V pipeline ────────────────────────────────────────────────
 pipe = WanImageToVideoPipeline.from_pretrained(
     MODEL_ID,
     transformer=WanTransformer3DModel.from_pretrained(
@@ -91,7 +87,6 @@ spaces.aoti_load(
     repo_id='cbensimon/WanTransformer3DModel-sm120-cu130-raa',
 )
 
-# ─── Load FLUX.2 Klein 4B upscaler pipeline ────────────────────────────────────
 print("Loading FLUX.2 Klein 4B model...")
 klein_pipe = Flux2KleinPipeline.from_pretrained(
     "black-forest-labs/FLUX.2-klein-4B",
@@ -99,13 +94,9 @@ klein_pipe = Flux2KleinPipeline.from_pretrained(
 ).to(device)
 print("FLUX.2 Klein 4B loaded successfully.")
 
-
-# ─── Prompts ───────────────────────────────────────────────────────────────────
 default_prompt_i2v = "make this image come alive, cinematic motion, smooth animation"
 default_negative_prompt = "色调艳丽, 过曝, 静态, 细节模糊不清, 字幕, 风格, 作品, 画作, 画面, 静止, 整体发灰, 最差质量, 低质量, JPEG压缩残留, 丑陋的, 残缺的, 多余的手指, 画得不好的手部, 画得不好的脸部, 畸形的, 毁容的, 形态畸形的肢体, 手指融合, 静止不动的画面, 杂乱的背景, 三条腿, 背景人很多, 倒着走"
 
-
-# ─── Image & frame helpers ─────────────────────────────────────────────────────
 def resize_image(image: Image.Image) -> Image.Image:
     """
     Resizes an image to fit within the model's constraints, preserving aspect ratio as much as possible.
@@ -219,7 +210,6 @@ def upscale_single_frame(frame: Image.Image) -> Image.Image:
     return result
 
 
-# ─── Video generation ──────────────────────────────────────────────────────────
 def get_duration(
     input_image,
     prompt,
@@ -285,7 +275,6 @@ def generate_video(
     return video_path, current_seed
 
 
-# ─── Full pipeline: generate → extract frames → upscale → return card data ───
 def generate_and_upscale(
     input_image,
     prompt,
@@ -317,7 +306,6 @@ def generate_and_upscale(
     if not frames:
         return video_path, current_seed, []
 
-    # Step 3: Upscale each frame
     upscaled_pairs = []
     for i, frame in enumerate(frames):
         progress(i / len(frames), desc=f"Upscaling frame {i+1}/{len(frames)}...")
@@ -348,7 +336,7 @@ def run_pipeline(
         - list of ImageSlider pairs for up to MAX_CARDS cards
         - list of upscaled image paths for download buttons
     """
-    MAX_CARDS = 10  # maximum frames we support in UI
+    MAX_CARDS = 10
 
     video_path, current_seed, upscaled_pairs = generate_and_upscale(
         input_image, prompt, steps, negative_prompt,
@@ -356,7 +344,6 @@ def run_pipeline(
         seed, randomize_seed, progress,
     )
 
-    # Build outputs for each card slot
     slider_outputs = []
     download_outputs = []
     visibility_outputs = []
@@ -373,8 +360,7 @@ def run_pipeline(
             slider_outputs.append(None)
             download_outputs.append(None)
             visibility_outputs.append(gr.update(visible=False))
-
-    # Flatten: video, seed, then interleaved (slider, download, visibility) per card
+            
     results = [video_path, current_seed]
     for i in range(MAX_CARDS):
         results.append(slider_outputs[i])
@@ -383,9 +369,6 @@ def run_pipeline(
 
     return results
 
-
-# ─── Build Gradio UI ──────────────────────────────────────────────────────────
-
 css = '''
 .upscale-card {
     border: 1px solid var(--border-color-primary);
@@ -414,7 +397,6 @@ with gr.Blocks(title="wan2.2-i2v-fast") as demo:
     )
 
     with gr.Row():
-        # ─── LEFT COLUMN: Inputs ──────────────────────────────────
         with gr.Column(scale=1):
             input_image_component = gr.Image(type="pil", label="Input Image", height=300)
             prompt_input = gr.Textbox(
@@ -452,7 +434,6 @@ with gr.Blocks(title="wan2.2-i2v-fast") as demo:
 
             generate_button = gr.Button("Generate Video & Upscale Frames", variant="primary")
 
-        # ─── RIGHT COLUMN: Video + Upscale Cards ──────────────────
         with gr.Column(scale=1):
             video_output = gr.Video(
                 label="Generated Video", autoplay=True, interactive=False
@@ -467,7 +448,6 @@ with gr.Blocks(title="wan2.2-i2v-fast") as demo:
                     "Click download to save the upscaled image.*"
                 )
     
-                # Create card slots dynamically
                 card_groups = []   # list of (group, slider, download_btn)
                 slider_components = []
                 download_components = []
@@ -490,7 +470,6 @@ with gr.Blocks(title="wan2.2-i2v-fast") as demo:
                     download_components.append(download_btn)
                     group_components.append(card_group)
 
-    # ─── Wire up inputs/outputs ────────────────────────────────────────────────
     ui_inputs = [
         input_image_component, prompt_input, steps_slider,
         negative_prompt_input, duration_seconds_input,
@@ -498,7 +477,6 @@ with gr.Blocks(title="wan2.2-i2v-fast") as demo:
         seed_input, randomize_seed_checkbox,
     ]
 
-    # Build output list: video, seed, then (slider, download, group_visibility) × MAX_CARDS
     ui_outputs = [video_output, seed_input]
     for i in range(MAX_CARDS):
         ui_outputs.append(slider_components[i])
@@ -511,21 +489,20 @@ with gr.Blocks(title="wan2.2-i2v-fast") as demo:
         outputs=ui_outputs,
     )
 
-    # ─── Examples ──────────────────────────────────────────────────────────────
     gr.Examples(
         examples=[
             [
-                "wan_i2v_input.JPG",
+                "example-file/wan_i2v_input.JPG",
                 "POV selfie video, white cat with sunglasses standing on surfboard, relaxed smile, tropical beach behind (clear water, green hills, blue sky with clouds). Surfboard tips, cat falls into ocean, camera plunges underwater with bubbles and sunlight beams. Brief underwater view of cat's face, then cat resurfaces, still filming selfie, playful summer vacation mood.",
                 4,
             ],
             [
-                "wan22_input_2.jpg",
+                "example-file/wan22_input_2.jpg",
                 "A sleek lunar vehicle glides into view from left to right, kicking up moon dust as astronauts in white spacesuits hop aboard with characteristic lunar bouncing movements. In the distant background, a VTOL craft descends straight down and lands silently on the surface. Throughout the entire scene, ethereal aurora borealis ribbons dance across the star-filled sky, casting shimmering curtains of green, blue, and purple light that bathe the lunar landscape in an otherworldly, magical glow.",
                 4,
             ],
             [
-                "kill_bill.jpeg",
+                "example-file/kill_bill.jpeg",
                 "Uma Thurman's character, Beatrix Kiddo, holds her razor-sharp katana blade steady in the cinematic lighting. Suddenly, the polished steel begins to soften and distort, like heated metal starting to lose its structural integrity. The blade's perfect edge slowly warps and droops, molten steel beginning to flow downward in silvery rivulets while maintaining its metallic sheen. The transformation starts subtly at first - a slight bend in the blade - then accelerates as the metal becomes increasingly fluid. The camera holds steady on her face as her piercing eyes gradually narrow, not with lethal focus, but with confusion and growing alarm as she watches her weapon dissolve before her eyes. Her breathing quickens slightly as she witnesses this impossible transformation. The melting intensifies, the katana's perfect form becoming increasingly abstract, dripping like liquid mercury from her grip. Molten droplets fall to the ground with soft metallic impacts. Her expression shifts from calm readiness to bewilderment and concern as her legendary instrument of vengeance literally liquefies in her hands, leaving her defenseless and disoriented.",
                 6,
             ],