feat: replace new model and delete unnecessary loading

app.py

@@ -47,15 +47,14 @@ MARKDOWN = \
             </h2> \
         
         <div style="display: flex; flex-wrap: wrap; justify-content: center; gap: 2rem; margin-bottom: 1rem;">
-            <!-- 第一行按钮 -->
             <a href="https://arxiv.org/abs/2505.21491" target="_blank"
             style="display: inline-flex; align-items: center; padding: 0.5rem 1rem; background-color: #f0f0f0; /* 浅灰色背景 */ color: #333; /* 深色文字 */ text-decoration: none; border-radius: 9999px; font-weight: 500; transition: background-color 0.3s;">
-                <span style="margin-right: 0.5rem;">📄</span> <!-- 使用文档图标 -->
+                <span style="margin-right: 0.5rem;">📄</span> 
                 <span>Paper</span>
             </a>
             <a href="https://github.com/UVA-Computer-Vision-Lab/FrameINO" target="_blank"
             style="display: inline-flex; align-items: center; padding: 0.5rem 1rem; background-color: #f0f0f0; color: #333; text-decoration: none; border-radius: 9999px; font-weight: 500; transition: background-color 0.3s;">
-                <span style="margin-right: 0.5rem;">💻</span> <!-- 使用电脑图标 -->
+                <span style="margin-right: 0.5rem;">💻</span> 
                 <span>GitHub</span>
             </a>
             <a href="https://uva-computer-vision-lab.github.io/Frame-In-N-Out" target="_blank"
@@ -87,7 +86,7 @@ MARKDOWN = \
     ❗️❗️❗️Instruction Steps:<br>
     1️⃣ Upload your first frame image. Set the size you want to resize to for <b>Resized Height for Input Image</b> and <b>Resized Width for Input Image</b>.  <br> 
     2️⃣ Set your <b>canvas top left</b> and <b>bottom right expansion</b>. The combined height and width should be the multiplier of 32. <br>
-        PLEASE ENSURE that <b>Canvas HEIGHT = 704</b> and <b>Canvas WIDTH = 1280</b> for the best performance (current training resolution). <br>
+        Recommend <b>Canvas HEIGHT = 704</b> and <b>Canvas WIDTH = 1280</b> for the best performance (Pre-trained training Resolution). <br>
     3️⃣ Click <b>Build the Canvas</b>.  <br>
     4️⃣ Provide the trajectory of the main object in the canvas by clicking on the <b>Expanded Canvas</b>. <br>
     5️⃣ Provide the ID reference image and its trajectory (optional). Also, write a detailed <b>text prompt</b>. <br>
@@ -122,48 +121,20 @@ if not os.path.exists("__assets__"):            # Check if the assets images exi
 
 
 
-
-######################################################## CogVideoX #################################################################
-
-# Path Setting
-model_code_name = "CogVideox"
-base_model_id = "zai-org/CogVideoX-5b-I2V"  
-transformer_ckpt_path = "uva-cv-lab/FrameINO_CogVideoX_Stage2_MotionINO_v1.0"
-
-# Load Model
-transformer = CogVideoXTransformer3DModel.from_pretrained(transformer_ckpt_path, torch_dtype=torch.float16)
-text_encoder = T5EncoderModel.from_pretrained(base_model_id, subfolder="text_encoder", torch_dtype=torch.float16)
-vae = AutoencoderKLCogVideoX.from_pretrained(base_model_id, subfolder="vae", torch_dtype=torch.float16)
-
-# Create pipeline and run inference
-pipe = CogVideoXImageToVideoPipeline.from_pretrained(
-            base_model_id,
-            text_encoder = text_encoder,
-            transformer = transformer,
-            vae = vae,
-            torch_dtype = torch.float16,
-        )
-pipe.enable_model_cpu_offload()
-
-#####################################################################################################################################
-
-
-
-
 ######################################################## Wan2.2 5B #################################################################
 
 # Path Setting
 model_code_name = "Wan"
 base_model_id = "Wan-AI/Wan2.2-TI2V-5B-Diffusers"  
-transformer_ckpt_path = "uva-cv-lab/FrameINO_Wan2.2_5B_Stage2_MotionINO_v1.5"
+transformer_ckpt_path = "uva-cv-lab/FrameINO_Wan2.2_5B_Stage2_MotionINO_v1.6"
 
 
-# Load model
+# Load Model
 print("Loading the model!")
 transformer = WanTransformer3DModel.from_pretrained(transformer_ckpt_path, torch_dtype=torch.float16)
 vae = AutoencoderKLWan.from_pretrained(base_model_id, subfolder="vae", torch_dtype=torch.float32)
 
-# Create the pipeline
+# Create the Pipeline
 print("Loading the pipeline!")
 pipe = WanImageToVideoPipeline.from_pretrained(base_model_id, transformer=transformer, vae=vae, torch_dtype=torch.bfloat16)
 pipe.to("cuda")