Remove SE

app.py

@@ -120,8 +120,8 @@ outputs_folder = './outputs/'
 os.makedirs(outputs_folder, exist_ok=True)
 
 input_image_debug_value = [None]
-input_video_debug_value = [None]
 end_image_debug_value = [None]
+input_video_debug_value = [None]
 prompt_debug_value = [None]
 total_second_length_debug_value = [None]
 
@@ -316,7 +316,7 @@ def set_mp4_comments_imageio_ffmpeg(input_file, comments):
         return False
 
 @torch.no_grad()
-def worker(input_image, end_image, image_position, prompts, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf, fps_number):
+def worker(input_image, end_image, image_position, prompts, n_prompt, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf, fps_number):
     def encode_prompt(prompt, n_prompt):
         llama_vec, clip_l_pooler = encode_prompt_conds(prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
 
@@ -585,275 +585,6 @@ def worker(input_image, end_image, image_position, prompts, n_prompt, seed, tota
     stream.output_queue.push(('end', None))
     return
 
-@torch.no_grad()
-def worker_start_end(input_image, end_image, image_position, prompts, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf, fps_number):
-    def encode_prompt(prompt, n_prompt):
-        llama_vec, clip_l_pooler = encode_prompt_conds(prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
-
-        if cfg == 1:
-            llama_vec_n, clip_l_pooler_n = torch.zeros_like(llama_vec), torch.zeros_like(clip_l_pooler)
-        else:
-            llama_vec_n, clip_l_pooler_n = encode_prompt_conds(n_prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
-
-        llama_vec, llama_attention_mask = crop_or_pad_yield_mask(llama_vec, length=512)
-        llama_vec_n, llama_attention_mask_n = crop_or_pad_yield_mask(llama_vec_n, length=512)
-
-        llama_vec = llama_vec.to(transformer.dtype)
-        llama_vec_n = llama_vec_n.to(transformer.dtype)
-        clip_l_pooler = clip_l_pooler.to(transformer.dtype)
-        clip_l_pooler_n = clip_l_pooler_n.to(transformer.dtype)
-        return [llama_vec, clip_l_pooler, llama_vec_n, clip_l_pooler_n, llama_attention_mask, llama_attention_mask_n]
-
-    total_latent_sections = (total_second_length * fps_number) / (latent_window_size * 4)
-    total_latent_sections = int(max(round(total_latent_sections), 1))
-
-    job_id = generate_timestamp()
-
-    stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Starting ...'))))
-
-    try:
-        # Clean GPU
-        if not high_vram:
-            unload_complete_models(
-                text_encoder, text_encoder_2, image_encoder, vae, transformer
-            )
-
-        # Text encoding
-
-        stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Text encoding ...'))))
-
-        if not high_vram:
-            fake_diffusers_current_device(text_encoder, gpu)  # since we only encode one text - that is one model move and one encode, offload is same time consumption since it is also one load and one encode.
-            load_model_as_complete(text_encoder_2, target_device=gpu)
-
-
-        prompt_parameters = []
-
-        for prompt_part in prompts[:total_latent_sections]:
-            prompt_parameters.append(encode_prompt(prompt_part, n_prompt))
-
-        # Clean GPU
-        if not high_vram:
-            unload_complete_models(
-                text_encoder, text_encoder_2
-            )
-
-        # Processing input image (start frame)
-        stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Processing start frame ...'))))
-
-        H, W, C = input_image.shape
-        height, width = find_nearest_bucket(H, W, resolution=640)
-        input_image_np = resize_and_center_crop(input_image, target_width=width, target_height=height)
-
-        Image.fromarray(input_image_np).save(os.path.join(outputs_folder, f'{job_id}_start.png'))
-
-        input_image_pt = torch.from_numpy(input_image_np).float() / 127.5 - 1
-        input_image_pt = input_image_pt.permute(2, 0, 1)[None, :, None]
-        
-        # Processing end image (if provided)
-        has_end_image = end_image is not None
-        if has_end_image:
-            stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Processing end frame ...'))))
-            
-            H_end, W_end, C_end = end_image.shape
-            end_image_np = resize_and_center_crop(end_image, target_width=width, target_height=height)
-            
-            Image.fromarray(end_image_np).save(os.path.join(outputs_folder, f'{job_id}_end.png'))
-            
-            end_image_pt = torch.from_numpy(end_image_np).float() / 127.5 - 1
-            end_image_pt = end_image_pt.permute(2, 0, 1)[None, :, None]
-
-        # VAE encoding
-        stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'VAE encoding ...'))))
-
-        if not high_vram:
-            load_model_as_complete(vae, target_device=gpu)
-
-        start_latent = vae_encode(input_image_pt, vae)
-        
-        if has_end_image:
-            end_latent = vae_encode(end_image_pt, vae)
-
-        # CLIP Vision
-        stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'CLIP Vision encoding ...'))))
-
-        if not high_vram:
-            load_model_as_complete(image_encoder, target_device=gpu)
-
-        image_encoder_output = hf_clip_vision_encode(input_image_np, feature_extractor, image_encoder)
-        image_encoder_last_hidden_state = image_encoder_output.last_hidden_state
-        
-        if has_end_image:
-            end_image_encoder_output = hf_clip_vision_encode(end_image_np, feature_extractor, image_encoder)
-            end_image_encoder_last_hidden_state = end_image_encoder_output.last_hidden_state
-            # Combine both image embeddings or use a weighted approach
-            image_encoder_last_hidden_state = (image_encoder_last_hidden_state + end_image_encoder_last_hidden_state) / 2
-
-        # Clean GPU
-        if not high_vram:
-            unload_complete_models(
-                image_encoder
-            )
-
-        # Dtype
-        image_encoder_last_hidden_state = image_encoder_last_hidden_state.to(transformer.dtype)
-
-        # Sampling
-        stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Start sampling ...'))))
-
-        rnd = torch.Generator("cpu").manual_seed(seed)
-        num_frames = latent_window_size * 4 - 3
-
-        history_latents = torch.zeros(size=(1, 16, 1 + 2 + 16, height // 8, width // 8), dtype=torch.float32, device=cpu)
-        start_latent = start_latent.to(history_latents)
-        if has_end_image:
-            end_latent = end_latent.to(history_latents)
-            
-        history_pixels = None
-        total_generated_latent_frames = 0
-
-        if total_latent_sections > 4:
-            # In theory the latent_paddings should follow the above sequence, but it seems that duplicating some
-            # items looks better than expanding it when total_latent_sections > 4
-            # One can try to remove below trick and just
-            # use `latent_paddings = list(reversed(range(total_latent_sections)))` to compare
-            latent_paddings = [3] + [2] * (total_latent_sections - 3) + [1, 0]
-        else:
-            # Convert an iterator to a list
-            latent_paddings = list(range(total_latent_sections - 1, -1, -1))
-
-        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) / fps_number) :.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
-
-        for latent_padding in latent_paddings:
-            is_last_section = latent_padding == 0
-            is_first_section = latent_padding == latent_paddings[0]
-            latent_padding_size = latent_padding * latent_window_size
-
-            if stream.input_queue.top() == 'end':
-                stream.output_queue.push(('end', None))
-                return
-
-            print(f'latent_padding_size = {latent_padding_size}, is_last_section = {is_last_section}, is_first_section = {is_first_section}')
-
-            if len(prompt_parameters) > 0:
-                [llama_vec, clip_l_pooler, llama_vec_n, clip_l_pooler_n, llama_attention_mask, llama_attention_mask_n] = prompt_parameters.pop(len(prompt_parameters) - 1)
-
-            indices = torch.arange(1 + latent_padding_size + latent_window_size + 1 + 2 + 16).unsqueeze(0)
-            clean_latent_indices_pre, blank_indices, latent_indices, clean_latent_indices_post, clean_latent_2x_indices, clean_latent_4x_indices = indices.split([1, latent_padding_size, latent_window_size, 1, 2, 16], dim=1)
-            clean_latent_indices = torch.cat([clean_latent_indices_pre, clean_latent_indices_post], dim=1)
-            
-            clean_latents_post, clean_latents_2x, clean_latents_4x = history_latents[:, :, :1 + 2 + 16, :, :].split([1, 2, 16], dim=2)
-
-            # Use end image latent for the first section if provided
-            if has_end_image and is_first_section:
-                clean_latents_post = end_latent
-
-            clean_latents = torch.cat([start_latent, clean_latents_post], dim=2)
-
-            if not high_vram:
-                unload_complete_models()
-                move_model_to_device_with_memory_preservation(transformer, target_device=gpu, preserved_memory_gb=gpu_memory_preservation)
-
-            if use_teacache:
-                transformer.initialize_teacache(enable_teacache=True, num_steps=steps)
-            else:
-                transformer.initialize_teacache(enable_teacache=False)
-
-            generated_latents = sample_hunyuan(
-                transformer=transformer,
-                sampler='unipc',
-                width=width,
-                height=height,
-                frames=num_frames,
-                real_guidance_scale=cfg,
-                distilled_guidance_scale=gs,
-                guidance_rescale=rs,
-                # shift=3.0,
-                num_inference_steps=steps,
-                generator=rnd,
-                prompt_embeds=llama_vec,
-                prompt_embeds_mask=llama_attention_mask,
-                prompt_poolers=clip_l_pooler,
-                negative_prompt_embeds=llama_vec_n,
-                negative_prompt_embeds_mask=llama_attention_mask_n,
-                negative_prompt_poolers=clip_l_pooler_n,
-                device=gpu,
-                dtype=torch.bfloat16,
-                image_embeddings=image_encoder_last_hidden_state,
-                latent_indices=latent_indices,
-                clean_latents=clean_latents,
-                clean_latent_indices=clean_latent_indices,
-                clean_latents_2x=clean_latents_2x,
-                clean_latent_2x_indices=clean_latent_2x_indices,
-                clean_latents_4x=clean_latents_4x,
-                clean_latent_4x_indices=clean_latent_4x_indices,
-                callback=callback,
-            )
-
-            if is_last_section:
-                generated_latents = torch.cat([start_latent.to(generated_latents), generated_latents], dim=2)
-
-            total_generated_latent_frames += int(generated_latents.shape[2])
-            history_latents = torch.cat([generated_latents.to(history_latents), history_latents], dim=2)
-
-            if not high_vram:
-                offload_model_from_device_for_memory_preservation(transformer, target_device=gpu, preserved_memory_gb=8)
-                load_model_as_complete(vae, target_device=gpu)
-
-            if history_pixels is None:
-                history_pixels = vae_decode(history_latents[:, :, :total_generated_latent_frames, :, :], vae).cpu()
-            else:
-                section_latent_frames = (latent_window_size * 2 + 1) if is_last_section else (latent_window_size * 2)
-                overlapped_frames = latent_window_size * 4 - 3
-
-                current_pixels = vae_decode(history_latents[:, :, :min(total_generated_latent_frames, section_latent_frames)], vae).cpu()
-                history_pixels = soft_append_bcthw(current_pixels, history_pixels, overlapped_frames)
-
-            if not high_vram:
-                unload_complete_models(vae)
-
-            if enable_preview or is_last_section:
-                output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
-                
-                save_bcthw_as_mp4(history_pixels, output_filename, fps=fps_number, crf=mp4_crf)
-                
-                print(f'Decoded. Pixel shape {history_pixels.shape}')
-                
-                stream.output_queue.push(('file', output_filename))
-
-            if is_last_section:
-                break
-    except:
-        traceback.print_exc()
-
-        if not high_vram:
-            unload_complete_models(
-                text_encoder, text_encoder_2, image_encoder, vae, transformer
-            )
-
-    stream.output_queue.push(('end', None))
-    return
-
 # 20250506 pftq: Modified worker to accept video input and clean frame count
 @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, enable_preview, use_teacache, no_resize, mp4_crf, num_clean_frames, vae_batch):
@@ -1144,7 +875,7 @@ def process_on_gpu(input_image, end_image, image_position, prompts, generation_m
     global stream
     stream = AsyncStream()
 
-    async_run(worker_start_end if generation_mode == "start_end" else worker, input_image, end_image, image_position, prompts, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf, fps_number)
+    async_run(worker, input_image, end_image, image_position, prompts, n_prompt, seed, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf, fps_number)
 
     output_filename = None
 
@@ -1186,12 +917,12 @@ def process(input_image,
             resolution=640,
             total_second_length=5,
             latent_window_size=9,
-            steps=30,
+            steps=25,
             cfg=1.0,
             gs=10.0,
             rs=0.0,
             gpu_memory_preservation=6,
-            enable_preview=False,
+            enable_preview=True,
             use_teacache=False,
             mp4_crf=16,
             fps_number=30
@@ -1199,13 +930,12 @@ def process(input_image,
     if auto_allocation:
         allocation_time = min(total_second_length * 60 * (1.5 if use_teacache else 3.0) * (1 + ((steps - 25) / 25))**2, 600)
 
-    if input_image_debug_value[0] is not None or end_image_debug_value[0] is not None or prompt_debug_value[0] is not None or total_second_length_debug_value[0] is not None:
+    if input_image_debug_value[0] is not None or prompt_debug_value[0] is not None or total_second_length_debug_value[0] is not None:
         input_image = input_image_debug_value[0]
-        end_image = end_image_debug_value[0]
         prompt = prompt_debug_value[0]
         total_second_length = total_second_length_debug_value[0]
         allocation_time = min(total_second_length_debug_value[0] * 60 * 100, 600)
-        input_image_debug_value[0] = end_image_debug_value[0] = input_video_debug_value[0] = prompt_debug_value[0] = total_second_length_debug_value[0] = None
+        input_image_debug_value[0] = prompt_debug_value[0] = total_second_length_debug_value[0] = None
 
     if torch.cuda.device_count() == 0:
         gr.Warning('Set this space to GPU config to make it work.')
@@ -1226,7 +956,7 @@ def process(input_image,
     yield gr.update(label="Previewed Frames"), None, '', '', gr.update(interactive=False), gr.update(interactive=True), gr.skip()
 
     yield from process_on_gpu(input_image,
-            end_image,
+                              end_image,
             image_position,
             prompts,
             generation_mode,
@@ -1297,7 +1027,7 @@ def process_video(input_video, prompt, n_prompt, randomize_seed, seed, auto_allo
         prompt = prompt_debug_value[0]
         total_second_length = total_second_length_debug_value[0]
         allocation_time = min(total_second_length_debug_value[0] * 60 * 100, 600)
-        input_image_debug_value[0] = end_image_debug_value[0] = input_video_debug_value[0] = prompt_debug_value[0] = total_second_length_debug_value[0] = None
+        input_video_debug_value[0] = prompt_debug_value[0] = total_second_length_debug_value[0] = None
 
     if torch.cuda.device_count() == 0:
         gr.Warning('Set this space to GPU config to make it work.')
@@ -1397,11 +1127,11 @@ with block:
     local_storage = gr.BrowserState(default_local_storage)
     with gr.Row():
         with gr.Column():
-            generation_mode = gr.Radio([["Text-to-Video", "text"], ["Image-to-Video", "image"], ["Start & end frames", "start_end"], ["Video Extension", "video"]], elem_id="generation-mode", label="Generation mode", value = "image")
+            generation_mode = gr.Radio([["Text-to-Video", "text"], ["Image-to-Video", "image"], ["Video Extension", "video"]], elem_id="generation-mode", label="Generation mode", value = "image")
             text_to_video_hint = gr.HTML("Text-to-Video badly works with a flash effect at the start. I discourage to use the Text-to-Video feature. You should rather generate an image with Flux and use Image-to-Video. You will save time.")
             input_image = gr.Image(sources='upload', type="numpy", label="Image", height=320)
-            end_image = gr.Image(sources='upload', type="numpy", label="End Frame (Optional)", height=320)
             image_position = gr.Slider(label="Image position", minimum=0, maximum=100, value=0, step=1, info='0=Video start; 100=Video end (lower quality)')
+            end_image = gr.Image(sources='upload', type="numpy", label="End Frame (Optional)", height=320)
             input_video = gr.Video(sources='upload', label="Input Video", height=320)
             timeless_prompt = gr.Textbox(label="Timeless prompt", info='Used on the whole duration of the generation', value='', placeholder="The creature starts to move, fast motion, fixed camera, focus motion, consistent arm, consistent position, mute colors, insanely detailed")
             prompt_number = gr.Slider(label="Timed prompt number", minimum=0, maximum=1000, value=0, step=1, info='Prompts will automatically appear')
@@ -1426,7 +1156,7 @@ with block:
                 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 and no break in brightness, but often makes hands and fingers slightly worse.')
 
-                n_prompt = gr.Textbox(label="Negative Prompt", value="Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", info='Requires using normal CFG (undistilled) instead of Distilled (set Distilled=1 and CFG > 1).')
+                n_prompt = gr.Textbox(label="Negative Prompt", value="Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", info='Requires using normal CFG (undistilled) instead of Distilled (set Distilled=1 and CFG > 1).')
 
                 fps_number = gr.Slider(label="Frame per seconds", info="The model is trained for 30 fps so other fps may generate weird results", minimum=10, maximum=60, value=30, step=1)
 
@@ -1476,7 +1206,6 @@ with block:
 
             with gr.Accordion("Debug", open=False):
                 input_image_debug = gr.Image(type="numpy", label="Image Debug", height=320)
-                end_image_debug = gr.Image(type="numpy", label="End Image Debug", height=320)
                 input_video_debug = gr.Video(sources='upload', label="Input Video Debug", height=320)
                 prompt_debug = gr.Textbox(label="Prompt Debug", value='')
                 total_second_length_debug = gr.Slider(label="Additional Video Length to Generate (seconds) Debug", minimum=1, maximum=120, value=1, step=0.1)
@@ -1488,6 +1217,7 @@ with block:
             progress_desc = gr.Markdown('', elem_classes='no-generating-animation')
             progress_bar = gr.HTML('', elem_classes='no-generating-animation')
 
+    # 20250506 pftq: Updated inputs to include num_clean_frames
     ips = [input_image, end_image, image_position, final_prompt, generation_mode, n_prompt, randomize_seed, seed, auto_allocation, allocation_time, resolution, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, enable_preview, use_teacache, mp4_crf, fps_number]
     ips_video = [input_video, final_prompt, n_prompt, randomize_seed, seed, auto_allocation, allocation_time, 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]
 
@@ -1501,7 +1231,7 @@ with block:
                         0, # image_position
                         "Overcrowed street in Japan, photorealistic, realistic, intricate details, 8k, insanely detailed",
                         "text", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1537,7 +1267,7 @@ with block:
                         0, # image_position
                         "A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The man talks and the woman listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks, the man stops talking and the man listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks and the man listens",
                         "image", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1561,7 +1291,7 @@ with block:
                         0, # image_position
                         "A dolphin emerges from the water, photorealistic, realistic, intricate details, 8k, insanely detailed",
                         "image", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1585,7 +1315,7 @@ with block:
                         1, # image_position
                         "A building starting to explode, photorealistic, realisitc, 8k, insanely detailed",
                         "image", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1609,7 +1339,7 @@ with block:
                         50, # image_position
                         "A building starting to explode, photorealistic, realisitc, 8k, insanely detailed",
                         "image", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1633,43 +1363,7 @@ with block:
                         100, # image_position
                         "A building starting to explode, photorealistic, realisitc, 8k, insanely detailed",
                         "image", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
-                        True, # randomize_seed
-                        42, # seed
-                        True, # auto_allocation
-                        180, # allocation_time
-                        672, # resolution
-                        1, # total_second_length
-                        9, # latent_window_size
-                        30, # steps
-                        1.0, # cfg
-                        10.0, # gs
-                        0.0, # rs
-                        6, # gpu_memory_preservation
-                        False, # enable_preview
-                        False, # use_teacache
-                        16, # mp4_crf
-                        30 # fps_number
-                    ],
-                ],
-            run_on_click = True,
-            fn = process,
-	        inputs = ips,
-            outputs = [result_video, preview_image, progress_desc, progress_bar, start_button, end_button, warning],
-            cache_examples = torch.cuda.device_count() > 0,
-        )
-
-    with gr.Row(elem_id="start_end_examples", visible=False):
-        gr.Examples(
-        label = "Examples from start and end frames",
-            examples = [
-                    [
-                        "./img_examples/Example2.webp", # input_image
-                        None, # end_image
-                        0, # image_position
-                        "A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The man talks and the woman listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks, the man stops talking and the man listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks and the man listens",
-                        "start_end", # generation_mode
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1702,7 +1396,7 @@ with block:
                     [
                         "./img_examples/Example1.mp4", # input_video
                         "View of the sea as far as the eye can see, from the seaside, a piece of land is barely visible on the horizon at the middle, the sky is radiant, reflections of the sun in the water, photorealistic, realistic, intricate details, 8k, insanely detailed",
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1726,7 +1420,7 @@ with block:
                     [
                         "./img_examples/Example1.mp4", # input_video
                         "View of the sea as far as the eye can see, from the seaside, a piece of land is barely visible on the horizon at the middle, the sky is radiant, reflections of the sun in the water, photorealistic, realistic, intricate details, 8k, insanely detailed",
-                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                        "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1764,7 +1458,7 @@ with block:
                     0, # image_position
                     "Overcrowed street in Japan, photorealistic, realistic, intricate details, 8k, insanely detailed",
                     "text", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -1799,7 +1493,7 @@ with block:
                     0, # image_position
                     "A dolphin emerges from the water, photorealistic, realistic, intricate details, 8k, insanely detailed",
                     "image", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -1823,7 +1517,7 @@ with block:
                     0, # image_position
                     "A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The man talks and the woman listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks, the man stops talking and the man listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks and the man listens",
                     "image", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -1847,7 +1541,7 @@ with block:
                     0, # image_position
                     "A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The woman talks and the man listens; A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The man talks, the woman stops talking and the woman listens A man on the left and a woman on the right face each other ready to start a conversation, large space between the persons, full view, full-length view, 3D, pixar, 3D render, CGI. The man talks and the woman listens",
                     "image", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -1871,7 +1565,7 @@ with block:
                     0, # image_position
                     "A boy is walking to the right, full view, full-length view, cartoon",
                     "image", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -1895,7 +1589,7 @@ with block:
                     100, # image_position
                     "A building starting to explode, photorealistic, realisitc, 8k, insanely detailed",
                     "image", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -1921,48 +1615,13 @@ with block:
         cache_examples = False,
     )
 
-    gr.Examples(
-        label = "🖼️ Examples from start and end frames",
-        examples = [
-                [
-                    "./img_examples/Example1.png", # input_image
-                    None, # end_image
-                    0, # image_position
-                    "A dolphin emerges from the water, photorealistic, realistic, intricate details, 8k, insanely detailed",
-                    "start_end", # generation_mode
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
-                    True, # randomize_seed
-                    42, # seed
-                    True, # auto_allocation
-                    180, # allocation_time
-                    672, # resolution
-                    1, # total_second_length
-                    9, # latent_window_size
-                    30, # steps
-                    1.0, # cfg
-                    10.0, # gs
-                    0.0, # rs
-                    6, # gpu_memory_preservation
-                    False, # enable_preview
-                    True, # use_teacache
-                    16, # mp4_crf
-                    30 # fps_number
-                ],
-            ],
-        run_on_click = True,
-        fn = process,
-	    inputs = ips,
-        outputs = [result_video, preview_image, progress_desc, progress_bar, start_button, end_button, warning],
-        cache_examples = False,
-    )
-
     gr.Examples(
         label = "🎥 Examples from video",
         examples = [
                 [
                     "./img_examples/Example1.mp4", # input_video
                     "View of the sea as far as the eye can see, from the seaside, a piece of land is barely visible on the horizon at the middle, the sky is radiant, reflections of the sun in the water, photorealistic, realistic, intricate details, 8k, insanely detailed",
-                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, poorly framed, blurred, blurry, over-smooth", # n_prompt
+                    "Missing arm, long hand, unrealistic position, impossible contortion, visible bone, muscle contraction, blurred, blurry, over-smooth", # n_prompt
                     True, # randomize_seed
                     42, # seed
                     True, # auto_allocation
@@ -2013,106 +1672,42 @@ with block:
     
     def handle_generation_mode_change(generation_mode_data):
         if generation_mode_data == "text":
-            return [
-            gr.update(visible = True),  # text_to_video_hint
-            gr.update(visible = False), # image_position
-            gr.update(visible = False), # input_image
-            gr.update(visible = False), # end_image
-            gr.update(visible = False), # input_video
-            gr.update(visible = True),  # start_button
-            gr.update(visible = False), # start_button_video
-            gr.update(visible = False), # no_resize
-            gr.update(visible = False), # batch
-            gr.update(visible = False), # num_clean_frames
-            gr.update(visible = False), # vae_batch
-            gr.update(visible = False), # prompt_hint
-            gr.update(visible = True)   # fps_number
-            ]
+            return [gr.update(visible = True), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = True), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = True)]
         elif generation_mode_data == "image":
-            return [
-            gr.update(visible = False), # text_to_video_hint
-            gr.update(visible = True),  # image_position
-            gr.update(visible = True),  # input_image
-            gr.update(visible = False), # end_image
-            gr.update(visible = False), # input_video
-            gr.update(visible = True),  # start_button
-            gr.update(visible = False), # start_button_video
-            gr.update(visible = False), # no_resize
-            gr.update(visible = False), # batch
-            gr.update(visible = False), # num_clean_frames
-            gr.update(visible = False), # vae_batch
-            gr.update(visible = False), # prompt_hint
-            gr.update(visible = True)   # fps_number
-            ]
-        elif generation_mode_data == "start_end":
-            return [
-            gr.update(visible = False), # text_to_video_hint
-            gr.update(visible = False), # image_position
-            gr.update(visible = True),  # input_image
-            gr.update(visible = True),  # end_image
-            gr.update(visible = False), # input_video
-            gr.update(visible = True),  # start_button
-            gr.update(visible = False), # start_button_video
-            gr.update(visible = False), # no_resize
-            gr.update(visible = False), # batch
-            gr.update(visible = False), # num_clean_frames
-            gr.update(visible = False), # vae_batch
-            gr.update(visible = False), # prompt_hint
-            gr.update(visible = True)   # fps_number
-            ]
+            return [gr.update(visible = False), gr.update(visible = True), gr.update(visible = True), gr.update(visible = False), gr.update(visible = True), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = True)]
         elif generation_mode_data == "video":
-            return [
-            gr.update(visible = False), # text_to_video_hint
-            gr.update(visible = False), # image_position
-            gr.update(visible = False), # input_image
-            gr.update(visible = False), # end_image
-            gr.update(visible = True),  # input_video
-            gr.update(visible = False), # start_button
-            gr.update(visible = True),  # start_button_video
-            gr.update(visible = True),  # no_resize
-            gr.update(visible = True),  # batch
-            gr.update(visible = True),  # num_clean_frames
-            gr.update(visible = True),  # vae_batch
-            gr.update(visible = True),  # prompt_hint
-            gr.update(visible = False)  # fps_number
-            ]
-
-    def handle_field_debug_change(input_image_debug_data, input_video_debug_data, end_image_debug_data, prompt_debug_data, total_second_length_debug_data):
+            return [gr.update(visible = False), gr.update(visible = False), gr.update(visible = False), gr.update(visible = True), gr.update(visible = False), gr.update(visible = True), gr.update(visible = True), gr.update(visible = True), gr.update(visible = True), gr.update(visible = True), gr.update(visible = True), gr.update(visible = False)]
+
+    
+    def handle_field_debug_change(input_image_debug_data, input_video_debug_data, prompt_debug_data, total_second_length_debug_data):
         print("handle_field_debug_change")
         input_image_debug_value[0] = input_image_debug_data
         input_video_debug_value[0] = input_video_debug_data
-        end_image_debug_value[0] = end_image_debug_data
         prompt_debug_value[0] = prompt_debug_data
         total_second_length_debug_value[0] = total_second_length_debug_data
         return []
 
     input_image_debug.upload(
         fn=handle_field_debug_change,
-        inputs=[input_image_debug, input_video_debug, end_image_debug, prompt_debug, total_second_length_debug],
+        inputs=[input_image_debug, input_video_debug, prompt_debug, total_second_length_debug],
         outputs=[]
     )
     
     input_video_debug.upload(
         fn=handle_field_debug_change,
-        inputs=[input_image_debug, input_video_debug, end_image_debug, prompt_debug, total_second_length_debug],
-        outputs=[]
-    )
-    
-    end_image_debug.upload(
-        fn=handle_field_debug_change,
-        inputs=[input_image_debug, input_video_debug, end_image_debug, prompt_debug, total_second_length_debug],
+        inputs=[input_image_debug, input_video_debug, prompt_debug, total_second_length_debug],
         outputs=[]
     )
     
     prompt_debug.change(
         fn=handle_field_debug_change,
-        inputs=[input_image_debug, input_video_debug, end_image_debug, prompt_debug, total_second_length_debug],
+        inputs=[input_image_debug, input_video_debug, prompt_debug, total_second_length_debug],
         outputs=[]
     )
     
     total_second_length_debug.change(
         fn=handle_field_debug_change,
-        inputs=[input_image_debug, input_video_debug, end_image_debug, prompt_debug, total_second_length_debug],
+        inputs=[input_image_debug, input_video_debug, prompt_debug, total_second_length_debug],
         outputs=[]
     )
     
@@ -2136,7 +1731,7 @@ with block:
     generation_mode.change(
         fn=handle_generation_mode_change,
         inputs=[generation_mode],
-        outputs=[text_to_video_hint, image_position, input_image, end_image, input_video, start_button, start_button_video, no_resize, batch, num_clean_frames, vae_batch, prompt_hint, fps_number]
+        outputs=[text_to_video_hint, image_position, input_image, input_video, start_button, start_button_video, no_resize, batch, num_clean_frames, vae_batch, prompt_hint, fps_number]
     )
 
     # Update display when the page loads
@@ -2144,7 +1739,7 @@ with block:
         fn=handle_generation_mode_change, inputs = [
         generation_mode
     ], outputs = [
-       text_to_video_hint, image_position, input_image, end_image, input_video, start_button, start_button_video, no_resize, batch, num_clean_frames, vae_batch, prompt_hint, fps_number
+       text_to_video_hint, image_position, input_image, input_video, start_button, start_button_video, no_resize, batch, num_clean_frames, vae_batch, prompt_hint, fps_number
     ]
     )