Add first_image2/first_image3 per-clip frame conditioning

app.py

@@ -347,6 +347,8 @@ def generate_video(
     remove_music: bool = False,
     prompt2: str = "",
     prompt3: str = "",
+    first_image2=None,
+    first_image3=None,
     progress=gr.Progress(track_tqdm=True),
 ):
     try:
@@ -441,6 +443,16 @@ def generate_video(
         if prompt2 and prompt2.strip():
             print(f"[Clip 2] Generating with prompt2: {prompt2[:50]}...")
             seed2 = current_seed + 1
+            if first_image2 is not None:
+                images2 = []
+                temp_first2_path = output_dir / f"temp_first2_{current_seed}.jpg"
+                if hasattr(first_image2, "save"):
+                    first_image2.save(temp_first2_path)
+                else:
+                    temp_first2_path = Path(first_image2)
+                images2.append(ImageConditioningInput(path=str(temp_first2_path), frame_idx=0, strength=1.0))
+            else:
+                images2 = images
             video_frames_iter2, audio2 = pipeline(
                 prompt=prompt2.strip(),
                 seed=seed2,
@@ -448,7 +460,7 @@ def generate_video(
                 width=int(width),
                 num_frames=num_frames,
                 frame_rate=frame_rate,
-                images=images,
+                images=images2,
                 enhance_prompt=enhance_prompt,
             )
             frames2 = [frame for frame in video_frames_iter2]
@@ -466,6 +478,16 @@ def generate_video(
         if prompt3 and prompt3.strip():
             print(f"[Clip 3] Generating with prompt3: {prompt3[:50]}...")
             seed3 = current_seed + 2
+            if first_image3 is not None:
+                images3 = []
+                temp_first3_path = output_dir / f"temp_first3_{current_seed}.jpg"
+                if hasattr(first_image3, "save"):
+                    first_image3.save(temp_first3_path)
+                else:
+                    temp_first3_path = Path(first_image3)
+                images3.append(ImageConditioningInput(path=str(temp_first3_path), frame_idx=0, strength=1.0))
+            else:
+                images3 = images
             video_frames_iter3, audio3 = pipeline(
                 prompt=prompt3.strip(),
                 seed=seed3,
@@ -473,7 +495,7 @@ def generate_video(
                 width=int(width),
                 num_frames=num_frames,
                 frame_rate=frame_rate,
-                images=images,
+                images=images3,
                 enhance_prompt=enhance_prompt,
             )
             frames3 = [frame for frame in video_frames_iter3]
@@ -545,6 +567,9 @@ with gr.Blocks(title="Element-16 Video", delete_cache=(3600, 7200)) as demo:  #
             output_video2 = gr.Video(label="Generated Video 2", autoplay=True)
             output_video3 = gr.Video(label="Generated Video 3", autoplay=True)
 
+    with gr.Row():
+        first_image2 = gr.Image(label="First Frame 2 (optional)", type="pil")
+        first_image3 = gr.Image(label="First Frame 3 (optional)", type="pil")
     with gr.Row():
         prompt2 = gr.Textbox(label="Prompt 2 (optional)", value="", lines=2)
         prompt3 = gr.Textbox(label="Prompt 3 (optional)", value="", lines=2)
@@ -597,7 +622,7 @@ with gr.Blocks(title="Element-16 Video", delete_cache=(3600, 7200)) as demo:  #
         inputs=[
             first_image, last_image, prompt, duration, enhance_prompt,
             seed, randomize_seed, height, width, negative_prompt, blur_amount, remove_music,
-            prompt2, prompt3,
+            prompt2, prompt3, first_image2, first_image3,
         ],
         outputs=[output_video, seed, output_video2, output_video3],
     )