imthanhlv commited on
Commit
a6a259b
·
verified ·
1 Parent(s): da87f25

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +2 -4
app.py CHANGED
@@ -42,18 +42,16 @@ prepare_latents_original = pipe.prepare_latents
42
 
43
  # pipe.prepare_latents = prepare_latents_loop
44
 
45
- # with the shift it will become step=0 0,1,2,3 -> step=1 1,2,3,0 -> step=2 2,3,0,1 -> step=3 3,0,1,2 -> step=4 0,1,2,3
46
- # so we only shift (N_FRAME-1)//8+1 times
47
-
48
  def modify_latents_callback(pipeline, step, timestep, callback_kwargs):
49
  print("Rolling latents on step", step)
50
  latents = callback_kwargs.get("latents")
51
  unpacked_latents = pipeline._unpack_latents(latents, (N_FRAME-1)//8+1, HEIGHT//32, WIDTH//32, 1, 1)
 
52
  modified_latents = torch.roll(unpacked_latents, shifts=1, dims=2)
53
  modified_latents = pipeline._pack_latents(modified_latents)
54
  return {"latents": modified_latents}
55
 
56
- @spaces.GPU(duration=120)
57
  def generate_gif(prompt, use_fixed_seed):
58
  seed = 0 if use_fixed_seed else torch.seed()
59
  generator = torch.Generator(device="cuda").manual_seed(seed)
 
42
 
43
  # pipe.prepare_latents = prepare_latents_loop
44
 
 
 
 
45
  def modify_latents_callback(pipeline, step, timestep, callback_kwargs):
46
  print("Rolling latents on step", step)
47
  latents = callback_kwargs.get("latents")
48
  unpacked_latents = pipeline._unpack_latents(latents, (N_FRAME-1)//8+1, HEIGHT//32, WIDTH//32, 1, 1)
49
+ # the frame order after each denoising step will be 0,1,2 -> 2,0,1 -> 1,2,0 -> 0,1,2 ...
50
  modified_latents = torch.roll(unpacked_latents, shifts=1, dims=2)
51
  modified_latents = pipeline._pack_latents(modified_latents)
52
  return {"latents": modified_latents}
53
 
54
+ @spaces.GPU(duration=140)
55
  def generate_gif(prompt, use_fixed_seed):
56
  seed = 0 if use_fixed_seed else torch.seed()
57
  generator = torch.Generator(device="cuda").manual_seed(seed)