add 512

app.py

@@ -17,22 +17,6 @@ from funcs import (
 )
 from utils.utils import instantiate_from_config
 
-ckpt_path_512 = "checkpoints/base_512_v1/model_512.ckpt"
-ckpt_dir_512 = "checkpoints/base_512_v1"
-os.makedirs(ckpt_dir_512, exist_ok=True)
-hf_hub_download(repo_id="MoonQiu/LongerCrafter", filename="model_512.ckpt", local_dir=ckpt_dir_512)
-
-# ckpt_path_1024 = "checkpoints/base_1024_v1/model.ckpt"
-# ckpt_dir_1024 = "checkpoints/base_1024_v1"
-# os.makedirs(ckpt_dir_1024, exist_ok=True)
-# hf_hub_download(repo_id="VideoCrafter/Text2Video-1024", filename="model.ckpt", local_dir=ckpt_dir_1024)
-
-# ckpt_path_256 = "checkpoints/base_256_v1/model_256.pth"
-# ckpt_dir_256 = "checkpoints/base_256_v1"
-# os.makedirs(ckpt_dir_256, exist_ok=True)
-# hf_hub_download(repo_id="MoonQiu/LongerCrafter", filename="model_256.pth", local_dir=ckpt_dir_256)
-
-
 def infer(prompt, output_size, seed, num_frames, ddim_steps, unconditional_guidance_scale, save_fps):
     window_size = 16
     window_stride = 4
@@ -40,11 +24,16 @@ def infer(prompt, output_size, seed, num_frames, ddim_steps, unconditional_guida
     if output_size == "320x512":
         width = 512 
         height = 320
+        ckpt_dir_512 = "checkpoints/base_512_v1"
+        ckpt_path_512 = "checkpoints/base_512_v1/model_512.ckpt"
         config_512 = "configs/inference_t2v_tconv512_v1.0_freenoise.yaml"
         config_512 = OmegaConf.load(config_512)
         model_config_512 = config_512.pop("model", OmegaConf.create())
         model_512 = instantiate_from_config(model_config_512)
         model_512 = model_512.cuda()
+        if not os.path.exists(ckpt_path_512):
+            os.makedirs(ckpt_dir_512, exist_ok=True)
+            hf_hub_download(repo_id="MoonQiu/LongerCrafter", filename="model_512.ckpt", local_dir=ckpt_dir_512)
         model_512 = load_model_checkpoint(model_512, ckpt_path_512)
         model_512.eval()
         model = model_512
@@ -52,11 +41,16 @@ def infer(prompt, output_size, seed, num_frames, ddim_steps, unconditional_guida
     elif output_size == "576x1024":
         width = 1024
         height = 576
+        ckpt_dir_1024 = "checkpoints/base_1024_v1"
+        ckpt_path_1024 = "checkpoints/base_1024_v1/model.ckpt"
         config_1024 = "configs/inference_t2v_1024_v1.0_freenoise.yaml"
         config_1024 = OmegaConf.load(config_1024)
         model_config_1024 = config_1024.pop("model", OmegaConf.create())
         model_1024 = instantiate_from_config(model_config_1024)
         model_1024 = model_1024.cuda()
+        if not os.path.exists(ckpt_path_1024):
+            os.makedirs(ckpt_dir_1024, exist_ok=True)
+            hf_hub_download(repo_id="VideoCrafter/Text2Video-1024", filename="model.ckpt", local_dir=ckpt_dir_1024)
         model_1024 = load_model_checkpoint(model_1024, ckpt_path_1024)
         model_1024.eval()
         model = model_1024
@@ -64,11 +58,16 @@ def infer(prompt, output_size, seed, num_frames, ddim_steps, unconditional_guida
     # elif output_size == "256x256":
     #     width = 256 
     #     height = 256
+    #     ckpt_dir_256 = "checkpoints/base_256_v1"
+    #     ckpt_path_256 = "checkpoints/base_256_v1/model_256.pth"
     #     config_256 = "configs/inference_t2v_tconv256_v1.0_freenoise.yaml"
     #     config_256 = OmegaConf.load(config_256)
     #     model_config_256 = config_256.pop("model", OmegaConf.create())
     #     model_256 = instantiate_from_config(model_config_256)
     #     model_256 = model_256.cuda()
+    #     if not os.path.exists(ckpt_path_256):
+    #         os.makedirs(ckpt_dir_256, exist_ok=True)
+    #         hf_hub_download(repo_id="MoonQiu/LongerCrafter", filename="model_256.pth", local_dir=ckpt_dir_256)
     #     model_256 = load_model_checkpoint(model_256, ckpt_path_256)
     #     model_256.eval()
     #     model = model_256
@@ -270,8 +269,7 @@ with gr.Blocks(css=css) as demo:
         with gr.Row():
             with gr.Accordion('FreeNoise Parameters (feel free to adjust these parameters based on your prompt): ', open=False):
                 with gr.Row():
-                    # output_size = gr.Dropdown(["576x1024"], value="576x1024", label="Output Size (around 900s for 576x1024)")
-                    output_size = gr.Dropdown(["320x512", "576x1024"], value="320x512", label="Output Size", info="576x1024 will cost around 900s")
+                    output_size = gr.Dropdown(["320x512", "576x1024"], value="320x512", label="Output Size", info="around 350s for 320x512, around 900s for 576x1024")
                 with gr.Row():
                     num_frames = gr.Slider(label='Frames (a multiple of 4)',
                              minimum=16,