Merge c5a6d87db7 into ec902046f6

README.md

@@ -36,7 +36,7 @@ In this repository, we present **Wan2.1**, a comprehensive and open suite of vid
 
 ## Community Works
 If your work has improved **Wan2.1** and you would like more people to see it, please inform us.
-- [Phantom](https://github.com/Phantom-video/Phantom) has developed a unified video generation framework for single and multi-subject references based on **Wan2.1-T2V-1.3B**. Please refer to [their examples](https://github.com/Phantom-video/Phantom).
+- [Phantom](https://github.com/Phantom-video/Phantom) has developed a unified video generation framework for single and multi-subject references based on both **Wan2.1-T2V-1.3B** and **Wan2.1-T2V-14B**. Please refer to [their examples](https://github.com/Phantom-video/Phantom).
 - [UniAnimate-DiT](https://github.com/ali-vilab/UniAnimate-DiT), based on **Wan2.1-14B-I2V**, has trained a Human image animation model and has open-sourced the inference and training code. Feel free to enjoy it!
 - [CFG-Zero](https://github.com/WeichenFan/CFG-Zero-star) enhances **Wan2.1** (covering both T2V and I2V models) from the perspective of CFG.
 - [TeaCache](https://github.com/ali-vilab/TeaCache) now supports **Wan2.1** acceleration, capable of increasing speed by approximately 2x. Feel free to give it a try!

wan/first_last_frame2video.py

@@ -13,6 +13,7 @@ import numpy as np
 import torch
 import torch.cuda.amp as amp
 import torch.distributed as dist
+import torchvision
 import torchvision.transforms.functional as TF
 from tqdm import tqdm
 
@@ -211,7 +212,12 @@ class WanFLF2V:
                 round(last_frame_size[1] * last_frame_resize_ratio),
             ]
             # 2. center crop
-            last_frame = TF.center_crop(last_frame, last_frame_size)
+            transform = torchvision.transforms.Compose([
+                torchvision.transforms.Resize((last_frame_size[0], last_frame_size[1])),
+                torchvision.transforms.CenterCrop((first_frame_size[0], first_frame_size[1]))
+            ])
+
+            last_frame = transform(last_frame)
 
         max_seq_len = ((F - 1) // self.vae_stride[0] + 1) * lat_h * lat_w // (
             self.patch_size[1] * self.patch_size[2])