Take resolution into account

app.py

@@ -120,8 +120,8 @@ outputs_folder = './outputs/'
 os.makedirs(outputs_folder, exist_ok=True)
 
 input_image_debug_value = [None]
-end_image_debug_value = [None]
 input_video_debug_value = [None]
+end_image_debug_value = [None]
 prompt_debug_value = [None]
 total_second_length_debug_value = [None]
 
@@ -586,7 +586,7 @@ def worker(input_image, end_image, image_position, prompts, n_prompt, seed, reso
     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 worker_start_end(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)
 
@@ -642,7 +642,7 @@ def worker_start_end(input_image, end_image, image_position, prompts, n_prompt,
         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)
+        height, width = find_nearest_bucket(H, W, resolution=resolution)
         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'))
@@ -1144,7 +1144,7 @@ def process_on_gpu(input_image, end_image, image_position, prompts, generation_m
     global stream
     stream = AsyncStream()
 
-    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)
+    async_run(worker_start_end if generation_mode == "start_end" else 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 +1186,12 @@ def process(input_image,
             resolution=640,
             total_second_length=5,
             latent_window_size=9,
-            steps=25,
+            steps=30,
             cfg=1.0,
             gs=10.0,
             rs=0.0,
             gpu_memory_preservation=6,
-            enable_preview=True,
+            enable_preview=False,
             use_teacache=False,
             mp4_crf=16,
             fps_number=30
@@ -1226,7 +1226,7 @@ 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 +1297,7 @@ 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_video_debug_value[0] = prompt_debug_value[0] = total_second_length_debug_value[0] = None
+        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
 
     if torch.cuda.device_count() == 0:
         gr.Warning('Set this space to GPU config to make it work.')
@@ -1400,8 +1400,8 @@ with block:
             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")
             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)
-            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)
+            image_position = gr.Slider(label="Image position", minimum=0, maximum=100, value=0, step=1, info='0=Video start; 100=Video end (lower quality)')
             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 +1426,7 @@ 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, 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, poorly framed, 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)
 
@@ -1488,7 +1488,6 @@ 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]
 
@@ -1502,7 +1501,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1538,7 +1537,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1562,7 +1561,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1586,7 +1585,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1610,7 +1609,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1634,7 +1633,43 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
                         True, # randomize_seed
                         42, # seed
                         True, # auto_allocation
@@ -1667,7 +1702,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1691,7 +1726,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                        "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
@@ -1729,7 +1764,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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
@@ -1764,7 +1799,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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
@@ -1788,7 +1823,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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
@@ -1812,7 +1847,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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
@@ -1836,7 +1871,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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
@@ -1860,7 +1895,7 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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
@@ -1886,13 +1921,48 @@ 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, blurred, blurry, over-smooth", # n_prompt
+                    "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