Try to use 2 models: one optimized for 25 f/s, another for 14 f/s

app.py

@@ -12,10 +12,15 @@ from PIL import Image
 import random
 import spaces
 
-pipe = StableVideoDiffusionPipeline.from_pretrained(
+fps25Pipe = StableVideoDiffusionPipeline.from_pretrained(
     "vdo/stable-video-diffusion-img2vid-xt-1-1", torch_dtype=torch.float16, variant="fp16"
 )
-pipe.to("cuda")
+fps25Pipe.to("cuda")
+
+fps14Pipe = StableVideoDiffusionPipeline.from_pretrained(
+    "stabilityai/stable-video-diffusion-img2vid", torch_dtype=torch.float16, variant="fp16"
+)
+fps14Pipe.to("cuda")
 
 max_64_bit_int = 2**63 - 1
 
@@ -44,7 +49,10 @@ def sample(
     base_count = len(glob(os.path.join(output_folder, "*.mp4")))
     video_path = os.path.join(output_folder, f"{base_count:06d}.mp4")
 
-    frames = pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
+    if 14 < fps_id:
+        frames = fps25Pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
+    else:
+        frames = fps14Pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
     export_to_video(frames, video_path, fps=fps_id)
     
     return video_path, gr.update(label="Generated frames in *." + frame_format + " format", format = frame_format, value = frames), seed