1inkusFace commited on
Commit
7413101
·
verified ·
1 Parent(s): 13ddf33

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +7 -7
app.py CHANGED
@@ -110,16 +110,16 @@ def generate(segment, image, prompt, size, guidance_scale, num_inference_steps,
110
  segment_timesteps = torch.from_numpy(timesteps_split_np[0]).to("cuda")
111
  num_channels_latents = pipe.transformer.config.in_channels
112
  num_channels_latents = int(num_channels_latents / 2)
113
- image = pipe.video_processor.preprocess(image, height=height, width=width).to(
114
  device, dtype=prompt_embeds.dtype
115
  )
116
  num_latent_frames = (frames - 1) // pipe.vae_scale_factor_temporal + 1
117
  latents = pipe.prepare_latents(
118
- batch_size=1, num_channels_latents=pipe.transformer.config.in_channels, height=height, width=width, num_frames=frames,
119
  dtype=torch.float32, device=device, generator=generator, latents=None,
120
  )
121
  image_latents = pipe.image_latents(
122
- image, batch_size, height, width, device, torch.float32, num_channels_latents, num_latent_frames
123
  )
124
  image_latents = image_latents.to(pipe.transformer.dtype)
125
  guidance = torch.tensor([guidance_scale] * latents.shape[0], dtype=transformer_dtype, device=device) * 1000.0
@@ -130,8 +130,8 @@ def generate(segment, image, prompt, size, guidance_scale, num_inference_steps,
130
  latents = state["intermediate_latents"].to("cuda", dtype=torch.bfloat16)
131
  guidance_scale = state["guidance_scale"]
132
  all_timesteps_cpu = state["all_timesteps"]
133
- height = state["height"]
134
- width = state["width"]
135
  pipe.scheduler.set_timesteps(len(all_timesteps_cpu), device=device)
136
  timesteps_split_np = np.array_split(all_timesteps_cpu.numpy(), 8)
137
  segment_timesteps = torch.from_numpy(timesteps_split_np[segment - 1]).to("cuda")
@@ -194,8 +194,8 @@ def generate(segment, image, prompt, size, guidance_scale, num_inference_steps,
194
  "seed": seed,
195
  "prompt": prompt, # Save originals for reference/verification
196
  "negative_prompt": negative_prompt,
197
- "height": height, # Save dimensions used
198
- "width": width
199
  }
200
  state_file = f"SkyReel_{segment}_{seed}.pt"
201
  torch.save(state, state_file)
 
110
  segment_timesteps = torch.from_numpy(timesteps_split_np[0]).to("cuda")
111
  num_channels_latents = pipe.transformer.config.in_channels
112
  num_channels_latents = int(num_channels_latents / 2)
113
+ image = pipe.video_processor.preprocess(image, height=size, width=size).to(
114
  device, dtype=prompt_embeds.dtype
115
  )
116
  num_latent_frames = (frames - 1) // pipe.vae_scale_factor_temporal + 1
117
  latents = pipe.prepare_latents(
118
+ batch_size=1, num_channels_latents=pipe.transformer.config.in_channels, height=size, width=size, num_frames=frames,
119
  dtype=torch.float32, device=device, generator=generator, latents=None,
120
  )
121
  image_latents = pipe.image_latents(
122
+ image, batch_size, size, size, device, torch.float32, num_channels_latents, num_latent_frames
123
  )
124
  image_latents = image_latents.to(pipe.transformer.dtype)
125
  guidance = torch.tensor([guidance_scale] * latents.shape[0], dtype=transformer_dtype, device=device) * 1000.0
 
130
  latents = state["intermediate_latents"].to("cuda", dtype=torch.bfloat16)
131
  guidance_scale = state["guidance_scale"]
132
  all_timesteps_cpu = state["all_timesteps"]
133
+ size = state["height"]
134
+ size = state["width"]
135
  pipe.scheduler.set_timesteps(len(all_timesteps_cpu), device=device)
136
  timesteps_split_np = np.array_split(all_timesteps_cpu.numpy(), 8)
137
  segment_timesteps = torch.from_numpy(timesteps_split_np[segment - 1]).to("cuda")
 
194
  "seed": seed,
195
  "prompt": prompt, # Save originals for reference/verification
196
  "negative_prompt": negative_prompt,
197
+ "height": size, # Save dimensions used
198
+ "width": size
199
  }
200
  state_file = f"SkyReel_{segment}_{seed}.pt"
201
  torch.save(state, state_file)