# Some parameters
n_points = 6 #@param
n_steps = 300 #@param
latents = torch.randn(n_points, 256)

# Loop through generating the frames
frames = []
for i in tqdm(range(n_steps)):
  p1 = max(0, int(n_points*i/n_steps))
  p2 = min(n_points, int(n_points*i/n_steps)+1)%n_points # so it wraps back to 0
  frac = (i-(p1*(n_steps/n_points))) / (n_steps/n_points)
  l = latents[p1]*(1-frac) + latents[p2]*frac
  im = model.G(l.unsqueeze(0)).clamp_(0., 1.)
  frame=(im[0].permute(1, 2, 0).detach().cpu().numpy()*255).astype(np.uint8)
  frames.append(frame)