enable_preview

app.py

@@ -305,7 +305,7 @@ def set_mp4_comments_imageio_ffmpeg(input_file, comments):
         return False
 
 @torch.no_grad()
-def worker(input_image, prompts, n_prompt, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, mp4_crf):
+def worker(input_image, prompts, n_prompt, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf):
     def encode_prompt(prompt, n_prompt):
         llama_vec, clip_l_pooler = encode_prompt_conds(prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
 
@@ -398,23 +398,27 @@ def worker(input_image, prompts, n_prompt, seed, resolution, total_second_length
         history_latents = torch.cat([history_latents, start_latent.to(history_latents)], dim=2)
         total_generated_latent_frames = 1
 
-        def callback(d):
-            preview = d['denoised']
-            preview = vae_decode_fake(preview)
-
-            preview = (preview * 255.0).detach().cpu().numpy().clip(0, 255).astype(np.uint8)
-            preview = einops.rearrange(preview, 'b c t h w -> (b h) (t w) c')
-
-            if stream.input_queue.top() == 'end':
-                stream.output_queue.push(('end', None))
-                raise KeyboardInterrupt('User ends the task.')
-
-            current_step = d['i'] + 1
-            percentage = int(100.0 * current_step / steps)
-            hint = f'Sampling {current_step}/{steps}'
-            desc = f'Total generated frames: {int(max(0, total_generated_latent_frames * 4 - 3))}, Video length: {max(0, (total_generated_latent_frames * 4 - 3) / 30) :.2f} seconds (FPS-30). The video is being extended now ...'
-            stream.output_queue.push(('progress', (preview, desc, make_progress_bar_html(percentage, hint))))
-            return
+        if enable_preview:
+            def callback(d):
+                preview = d['denoised']
+                preview = vae_decode_fake(preview)
+			    
+                preview = (preview * 255.0).detach().cpu().numpy().clip(0, 255).astype(np.uint8)
+                preview = einops.rearrange(preview, 'b c t h w -> (b h) (t w) c')
+			    
+                if stream.input_queue.top() == 'end':
+                    stream.output_queue.push(('end', None))
+                    raise KeyboardInterrupt('User ends the task.')
+			    
+                current_step = d['i'] + 1
+                percentage = int(100.0 * current_step / steps)
+                hint = f'Sampling {current_step}/{steps}'
+                desc = f'Total generated frames: {int(max(0, total_generated_latent_frames * 4 - 3))}, Video length: {max(0, (total_generated_latent_frames * 4 - 3) / 30) :.2f} seconds (FPS-30), Resolution: {height}px * {width}px. The video is being extended now ...'
+                stream.output_queue.push(('progress', (preview, desc, make_progress_bar_html(percentage, hint))))
+                return
+        else:
+            def callback(d):
+                return
 
         indices = torch.arange(0, sum([1, 16, 2, 1, latent_window_size])).unsqueeze(0)
         clean_latent_indices_start, clean_latent_4x_indices, clean_latent_2x_indices, clean_latent_1x_indices, latent_indices = indices.split([1, 16, 2, 1, latent_window_size], dim=1)
@@ -494,13 +498,14 @@ def worker(input_image, prompts, n_prompt, seed, resolution, total_second_length
             if not high_vram:
                 unload_complete_models()
 
-            output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
+            if enable_preview or section_index == total_latent_sections - 1:
+                output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
 
-            save_bcthw_as_mp4(history_pixels, output_filename, fps=30, crf=mp4_crf)
+                save_bcthw_as_mp4(history_pixels, output_filename, fps=30, crf=mp4_crf)
 
-            print(f'Decoded. Current latent shape {real_history_latents.shape}; pixel shape {history_pixels.shape}')
+                print(f'Decoded. Current latent shape {real_history_latents.shape}; pixel shape {history_pixels.shape}')
 
-            stream.output_queue.push(('file', output_filename))
+                stream.output_queue.push(('file', output_filename))
     except:
         traceback.print_exc()
 
@@ -512,12 +517,12 @@ def worker(input_image, prompts, n_prompt, seed, resolution, total_second_length
     stream.output_queue.push(('end', None))
     return
 
-def get_duration(input_image, prompt, generation_mode, n_prompt, randomize_seed, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, mp4_crf):
+def get_duration(input_image, prompt, generation_mode, n_prompt, randomize_seed, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf):
     global total_second_length_debug_value
 
     if total_second_length_debug_value is not None:
         return min(total_second_length_debug_value * 60 * 10, 600)
-    return total_second_length * 60 * (0.7 if use_teacache else 1.3) * (2**((resolution - 640) / 640))
+    return total_second_length * 60 * (0.7 if use_teacache else 1.3) * (2**((resolution - 640) / 640)) * (1 + ((steps - 25) / 100))
 
 
 @spaces.GPU(duration=get_duration)
@@ -534,6 +539,7 @@ def process(input_image, prompt,
             gs=10.0,
             rs=0.0,
             gpu_memory_preservation=6,
+            enable_preview=True,
             use_teacache=False,
             mp4_crf=16
            ):
@@ -565,7 +571,7 @@ def process(input_image, prompt,
 
     stream = AsyncStream()
 
-    async_run(worker, input_image, prompts, n_prompt, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, mp4_crf)
+    async_run(worker, input_image, prompts, n_prompt, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf)
 
     output_filename = None
 
@@ -587,7 +593,7 @@ def process(input_image, prompt,
 # 20250506 pftq: Modified worker to accept video input and clean frame count
 @spaces.GPU()
 @torch.no_grad()
-def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
+def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
     def encode_prompt(prompt, n_prompt):
         llama_vec, clip_l_pooler = encode_prompt_conds(prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
 
@@ -647,23 +653,27 @@ def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_
         total_latent_sections = (total_second_length * fps) / (latent_window_size * 4)
         total_latent_sections = int(max(round(total_latent_sections), 1))
 
-        def callback(d):
-            preview = d['denoised']
-            preview = vae_decode_fake(preview)
-
-            preview = (preview * 255.0).detach().cpu().numpy().clip(0, 255).astype(np.uint8)
-            preview = einops.rearrange(preview, 'b c t h w -> (b h) (t w) c')
-
-            if stream.input_queue.top() == 'end':
-                stream.output_queue.push(('end', None))
-                raise KeyboardInterrupt('User ends the task.')
-
-            current_step = d['i'] + 1
-            percentage = int(100.0 * current_step / steps)
-            hint = f'Sampling {current_step}/{steps}'
-            desc = f'Total frames: {int(max(0, total_generated_latent_frames * 4 - 3))}, Video length: {max(0, (total_generated_latent_frames * 4 - 3) / fps) :.2f} seconds (FPS-{fps}), Resolution: {height}px * {width}px, Seed: {seed}, Video {idx+1} of {batch}. The video is generating part {section_index+1} of {total_latent_sections}...'
-            stream.output_queue.push(('progress', (preview, desc, make_progress_bar_html(percentage, hint))))
-            return
+        if enable_preview:
+            def callback(d):
+                preview = d['denoised']
+                preview = vae_decode_fake(preview)
+			    
+                preview = (preview * 255.0).detach().cpu().numpy().clip(0, 255).astype(np.uint8)
+                preview = einops.rearrange(preview, 'b c t h w -> (b h) (t w) c')
+			    
+                if stream.input_queue.top() == 'end':
+                    stream.output_queue.push(('end', None))
+                    raise KeyboardInterrupt('User ends the task.')
+			    
+                current_step = d['i'] + 1
+                percentage = int(100.0 * current_step / steps)
+                hint = f'Sampling {current_step}/{steps}'
+                desc = f'Total frames: {int(max(0, total_generated_latent_frames * 4 - 3))}, Video length: {max(0, (total_generated_latent_frames * 4 - 3) / fps) :.2f} seconds (FPS-{fps}), Resolution: {height}px * {width}px, Seed: {seed}, Video {idx+1} of {batch}. The video is generating part {section_index+1} of {total_latent_sections}...'
+                stream.output_queue.push(('progress', (preview, desc, make_progress_bar_html(percentage, hint))))
+                return
+        else:
+            def callback(d):
+                return
 
         for idx in range(batch):
             if batch > 1:
@@ -813,27 +823,28 @@ def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_
                 if not high_vram:
                     unload_complete_models()
 
-                output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
-
-                # 20250506 pftq: Use input video FPS for output
-                save_bcthw_as_mp4(history_pixels, output_filename, fps=fps, crf=mp4_crf)
-                print(f"Latest video saved: {output_filename}")
-                # 20250508 pftq: Save prompt to mp4 metadata comments
-                set_mp4_comments_imageio_ffmpeg(output_filename, f"Prompt: {prompts} | Negative Prompt: {n_prompt}");
-                print(f"Prompt saved to mp4 metadata comments: {output_filename}")
-
-                # 20250506 pftq: Clean up previous partial files
-                if previous_video is not None and os.path.exists(previous_video):
-                    try:
-                        os.remove(previous_video)
-                        print(f"Previous partial video deleted: {previous_video}")
-                    except Exception as e:
-                        print(f"Error deleting previous partial video {previous_video}: {e}")
-                previous_video = output_filename
-
-                print(f'Decoded. Current latent shape {real_history_latents.shape}; pixel shape {history_pixels.shape}')
-
-                stream.output_queue.push(('file', output_filename))
+                if enable_preview or section_index == total_latent_sections - 1:
+                    output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
+
+                    # 20250506 pftq: Use input video FPS for output
+                    save_bcthw_as_mp4(history_pixels, output_filename, fps=fps, crf=mp4_crf)
+                    print(f"Latest video saved: {output_filename}")
+                    # 20250508 pftq: Save prompt to mp4 metadata comments
+                    set_mp4_comments_imageio_ffmpeg(output_filename, f"Prompt: {prompts} | Negative Prompt: {n_prompt}");
+                    print(f"Prompt saved to mp4 metadata comments: {output_filename}")
+				    
+                    # 20250506 pftq: Clean up previous partial files
+                    if previous_video is not None and os.path.exists(previous_video):
+                        try:
+                            os.remove(previous_video)
+                            print(f"Previous partial video deleted: {previous_video}")
+                        except Exception as e:
+                            print(f"Error deleting previous partial video {previous_video}: {e}")
+                    previous_video = output_filename
+				    
+                    print(f'Decoded. Current latent shape {real_history_latents.shape}; pixel shape {history_pixels.shape}')
+				    
+                    stream.output_queue.push(('file', output_filename))
 
             seed = (seed + 1) % np.iinfo(np.int32).max
 
@@ -848,15 +859,15 @@ def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_
     stream.output_queue.push(('end', None))
     return
 
-def get_duration_video(input_video, prompt, n_prompt, randomize_seed, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
+def get_duration_video(input_video, prompt, n_prompt, randomize_seed, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
     global total_second_length_debug_value
     if total_second_length_debug_value is not None:
         return min(total_second_length_debug_value * 60 * 10, 600)
-    return total_second_length * 60 * (0.7 if use_teacache else 2) * (2**((resolution - 640) / 640))
+    return total_second_length * 60 * (0.7 if use_teacache else 2) * (2**((resolution - 640) / 640)) * (1 + ((steps - 25) / 100))
 
 # 20250506 pftq: Modified process to pass clean frame count, etc from video_encode
 @spaces.GPU(duration=get_duration_video)
-def process_video(input_video, prompt, n_prompt, randomize_seed, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
+def process_video(input_video, prompt, n_prompt, randomize_seed, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
     global stream, high_vram, input_video_debug_value, prompt_debug_value, total_second_length_debug_value
 
     if torch.cuda.device_count() == 0:
@@ -896,7 +907,7 @@ def process_video(input_video, prompt, n_prompt, randomize_seed, seed, batch, re
     stream = AsyncStream()
 
     # 20250506 pftq: Pass num_clean_frames, vae_batch, etc
-    async_run(worker_video, input_video, prompts, n_prompt, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch)
+    async_run(worker_video, input_video, prompts, n_prompt, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch)
 
     output_filename = None
 
@@ -1007,6 +1018,7 @@ with block:
                 end_button = gr.Button(elem_id="end-button", value="End Generation", variant="stop", interactive=False)
 
             with gr.Accordion("Advanced settings", open=False):
+                enable_preview = gr.Checkbox(label='Enable preview', value=True, info='Display a preview around each second generated but it costs 2 sec. for each second generated.')
                 use_teacache = gr.Checkbox(label='Use TeaCache', value=False, info='Faster speed, but often makes hands and fingers slightly worse.')
 
                 n_prompt = gr.Textbox(label="Negative Prompt", value="Missing arm, unrealistic position, impossible contortion, blurred, blurry", info='Requires using normal CFG (undistilled) instead of Distilled (set Distilled=1 and CFG > 1).')
@@ -1065,8 +1077,8 @@ with block:
             progress_bar = gr.HTML('', elem_classes='no-generating-animation')
 
     # 20250506 pftq: Updated inputs to include num_clean_frames
-    ips = [input_image, final_prompt, generation_mode, n_prompt, randomize_seed, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, mp4_crf]
-    ips_video = [input_video, final_prompt, n_prompt, randomize_seed, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch]
+    ips = [input_image, final_prompt, generation_mode, n_prompt, randomize_seed, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf]
+    ips_video = [input_video, final_prompt, n_prompt, randomize_seed, seed, batch, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch]
 
     def save_preferences(preferences, value):
         preferences["generation-mode"] = value
@@ -1123,6 +1135,7 @@ with block:
                     10.0, # gs
                     0.0, # rs
                     6, # gpu_memory_preservation
+                    False, # enable_preview
                     False, # use_teacache
                     16 # mp4_crf
                 ],
@@ -1141,6 +1154,7 @@ with block:
                     10.0, # gs
                     0.0, # rs
                     6, # gpu_memory_preservation
+                    False, # enable_preview
                     False, # use_teacache
                     16 # mp4_crf
                 ],
@@ -1170,6 +1184,7 @@ with block:
                     10.0, # gs
                     0.0, # rs
                     6, # gpu_memory_preservation
+                    False, # enable_preview
                     False, # use_teacache
                     False, # no_resize
                     16, # mp4_crf
@@ -1191,6 +1206,7 @@ with block:
                     10.0, # gs
                     0.0, # rs
                     6, # gpu_memory_preservation
+                    False, # enable_preview
                     False, # use_teacache
                     False, # no_resize
                     16, # mp4_crf
@@ -1222,6 +1238,7 @@ with block:
                     10.0, # gs
                     0.0, # rs
                     6, # gpu_memory_preservation
+                    False, # enable_preview
                     False, # use_teacache
                     16 # mp4_crf
                 ]
@@ -1250,6 +1267,7 @@ with block:
                     10.0, # gs
                     0.0, # rs
                     6, # gpu_memory_preservation
+                    False, # enable_preview
                     False, # use_teacache
                     False, # no_resize
                     16, # mp4_crf