Better images to video

.gitignore

@@ -1,2 +1,3 @@
 __pycache__/
-SaveText/
+SaveText/
+API_example/

README.md

@@ -1,4 +1,4 @@
-# 🔗 Comfyui : Bjornulf_custom_nodes v0.3 🔗
+# 🔗 Comfyui : Bjornulf_custom_nodes v0.4 🔗
 
 # Dependencies
 
@@ -9,6 +9,7 @@
 - **v0.2 Ollama**: Improve ollama node with system prompt + model selection.
 - **v0.3 Save Image to Folder**: Add a new node : Save image to a chosen folder.
 - **v0.3 Save Images**: Add comfyui Metadata / workflow to all my image-related nodes.
+- **v0.3 Images to video**: Support transparency with webm format. As well as an audio stream.
 
 # 📝 Nodes descriptions
 
@@ -137,4 +138,4 @@ Create a ping-pong effect from a list of images (from a video) by reversing the
 
 **Description:**  
 Combine a sequence of images into a video file.  
-❓ I made this node because it supports transparency with webm format. (Good for rembg)
+❓ I made this node because it supports transparency with webm format. (Need for rembg)

__init__.py

@@ -1,4 +1,4 @@
-from .create_video import imgs2vid
+from .images_to_video import imagesToVideo
 from .write_text import WriteText
 from .write_image_environment import WriteImageEnvironment
 from .write_image_characters import WriteImageCharacters
@@ -49,7 +49,7 @@ NODE_CLASS_MAPPINGS = {
     "Bjornulf_LoopTexts": LoopTexts,
     "Bjornulf_RandomTexts": RandomTexts,
     "Bjornulf_RandomModelClipVae": RandomModelClipVae,
-    "Bjornulf_imgs2vid": imgs2vid,
+    "Bjornulf_imagesToVideo": imagesToVideo,
     "Bjornulf_VideoPingPong": VideoPingPong,
     "Bjornulf_LoopFloat": LoopFloat,
     "Bjornulf_LoopInteger": LoopInteger,
@@ -81,7 +81,7 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "Bjornulf_LoopTexts": "♻ Loop (Texts)",
     "Bjornulf_RandomTexts": "🎲 Random (Texts)",
     "Bjornulf_RandomModelClipVae": "🎲 Random (Model+Clip+Vae)",
-    "Bjornulf_imgs2vid": "📹 imgs2vid (FFmpeg)",
+    "Bjornulf_imagesToVideo": "📹 images to video (FFmpeg)",
     "Bjornulf_VideoPingPong": "📹 video PingPong",
     "Bjornulf_LoopFloat": "♻ Loop (Float)",
     "Bjornulf_LoopInteger": "♻ Loop (Integer)",

create_video.py

@@ -1,91 +0,0 @@
-import os
-import numpy as np
-import torch
-import subprocess
-from PIL import Image
-
-class imgs2vid:
-    @classmethod
-    def INPUT_TYPES(cls):
-        return {
-            "required": {
-                "images": ("IMAGE",),
-                "fps": ("INT", {"default": 30, "min": 1, "max": 60}),
-                "video_name_NO_format": ("STRING", {"default": "output"}),
-                "format": (["mp4", "webm"],),
-                "audio_path": ("STRING", {"default": "/home/umen/6sec.wav"}),  # New audio input
-            },
-        }
-
-    RETURN_TYPES = ("STRING",)
-    RETURN_NAMES = ("comment",)
-    FUNCTION = "create_video"
-    OUTPUT_NODE = True
-    CATEGORY = "Bjornulf"
-
-    def create_video(self, images, fps, video_name_NO_format, format, audio_path):
-        # Remove any existing extension
-        video_name_NO_format = os.path.splitext(video_name_NO_format)[0]
-        # Add the correct extension
-        output_file = f"{video_name_NO_format}.{format}"
-        temp_dir = "temp_images"
-        os.makedirs(temp_dir, exist_ok=True)
-        # Ensure the output directory exists
-        os.makedirs(os.path.dirname(output_file) if os.path.dirname(output_file) else ".", exist_ok=True)
-
-        # Save the tensor images as PNG files
-        for i, img_tensor in enumerate(images):
-            img = Image.fromarray((img_tensor.cpu().numpy() * 255).astype(np.uint8))
-            if format == "webm":
-                img = img.convert("RGBA")  # Ensure alpha channel for WebM
-            img.save(os.path.join(temp_dir, f"frame_{i:04d}.png"))
-
-        # Construct the FFmpeg command based on the selected format
-        if format == "mp4":
-            ffmpeg_cmd = [
-                "ffmpeg",
-                "-y",
-                "-framerate", str(fps),
-                "-i", os.path.join(temp_dir, "frame_%04d.png"),
-                "-i", str(audio_path),
-                "-crf", "19",
-                "-c:v", "libx264",
-                "-pix_fmt", "yuv420p",
-                output_file
-            ]
-            comment = "MP4 format: Widely compatible, efficient compression, no transparency support."
-        elif format == "webm":
-            ffmpeg_cmd = [
-                "ffmpeg",
-                "-y",
-                "-framerate", str(fps),
-                "-i", os.path.join(temp_dir, "frame_%04d.png"),
-                "-i", str(audio_path),
-                "-crf", "19",
-                "-c:v", "libvpx",
-                "-b:v", "1M",  # Set video bitrate
-                "-auto-alt-ref", "0",  # Disable auto alt ref
-                "-c:a", "libvorbis",
-                "-pix_fmt", "yuva420p",
-                "-shortest",
-                output_file
-            ]
-            comment = "WebM format: Supports transparency, open format, smaller file size, but less compatible than MP4."
-
-        # Run FFmpeg
-        try:
-            subprocess.run(ffmpeg_cmd, check=True)
-            print(f"Video created successfully: {output_file}")
-        except subprocess.CalledProcessError as e:
-            print(f"Error creating video: {e}")
-        finally:
-            # Clean up temporary files
-            for file in os.listdir(temp_dir):
-                os.remove(os.path.join(temp_dir, file))
-            os.rmdir(temp_dir)
-
-        return (comment,)
-
-# Example usage
-# images = [torch.rand(256, 256, 3) for _ in range(10)]  # Replace with actual image tensors
-# imgs2vid().create_video(images, 30, "output", "webm", "/home/

images_to_video.py

@@ -0,0 +1,111 @@
+import os
+import numpy as np
+import torch
+import subprocess
+from PIL import Image
+import soundfile as sf
+
+class imagesToVideo:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "images": ("IMAGE",),
+                "fps": ("INT", {"default": 24, "min": 1, "max": 60}),
+                "name_prefix": ("STRING", {"default": "output/imgs2video/me"}),
+                "format": (["mp4", "webm"],),
+                "crf": ("INT", {"default": 19, "min": 0, "max": 63}),
+            },
+            "optional": {
+                "audio": ("AUDIO",),
+            },
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("comment",)
+    FUNCTION = "image_to_video"
+    OUTPUT_NODE = True
+    CATEGORY = "Bjornulf"
+
+    def image_to_video(self, images, fps, name_prefix, format, crf, audio=None):
+        # Remove any existing extension
+        name_prefix = os.path.splitext(name_prefix)[0]
+        # Add the correct extension
+        output_file = f"{name_prefix}.{format}"
+        temp_dir = "temp_images_imgs2video"
+        #Clean up temp dir
+        if os.path.exists(temp_dir) and os.path.isdir(temp_dir):
+            for file in os.listdir(temp_dir):
+                os.remove(os.path.join(temp_dir, file))
+            os.rmdir(temp_dir)
+        
+        os.makedirs(temp_dir, exist_ok=True)
+        # Ensure the output directory exists
+        os.makedirs(os.path.dirname(output_file) if os.path.dirname(output_file) else ".", exist_ok=True)
+
+        # Save the tensor images as PNG files
+        for i, img_tensor in enumerate(images):
+            img = Image.fromarray((img_tensor.cpu().numpy() * 255).astype(np.uint8))
+            if format == "webm":
+                img = img.convert("RGBA")  # Ensure alpha channel for WebM
+            img.save(os.path.join(temp_dir, f"frame_{i:04d}.png"))
+
+        # Handle audio
+        temp_audio_file = None
+        if audio is not None:
+            temp_audio_file = os.path.join(temp_dir, "temp_audio.wav")
+            waveform = audio['waveform'].squeeze().numpy()
+            sample_rate = audio['sample_rate']
+            sf.write(temp_audio_file, waveform, sample_rate)
+
+        # Construct the FFmpeg command based on the selected format
+        ffmpeg_cmd = [
+            "ffmpeg",
+            "-y",
+            "-framerate", str(fps),
+            "-i", os.path.join(temp_dir, "frame_%04d.png"),
+        ]
+
+        if temp_audio_file:
+            ffmpeg_cmd.extend(["-i", temp_audio_file])
+
+        if format == "mp4":
+            ffmpeg_cmd.extend([
+                "-filter_complex", "[0:v]scale=iw:ih,format=rgba,split[s0][s1];[s0]lutrgb=r=0:g=0:b=0:a=0[transparent];[transparent][s1]overlay",
+                "-crf", str(crf),
+                "-c:v", "libx264",
+                "-pix_fmt", "yuv420p",
+            ])
+            comment = "MP4 format: Widely compatible, efficient compression, no transparency support."
+        elif format == "webm":
+            # Fake transparency bug/feature with Inspyre.
+            # Code to fix tat : creates a fully transparent background and then overlays your image on top of it, which forces the transparency to be preserved... wth is this guys?
+            ffmpeg_cmd.extend([
+                "-filter_complex", "[0:v]scale=iw:ih,format=rgba,split[s0][s1];[s0]lutrgb=r=0:g=0:b=0:a=0[transparent];[transparent][s1]overlay",
+                "-c:v", "libvpx-vp9",
+                "-pix_fmt", "yuva420p",
+                "-b:v", "0",
+                "-crf", str(crf),
+                "-auto-alt-ref", "0",
+            ])
+            comment = "WebM format: Supports transparency, open format, smaller file size, but less compatible than MP4."
+
+        if temp_audio_file:
+            ffmpeg_cmd.extend(["-c:a", "libvorbis", "-shortest"])
+
+        ffmpeg_cmd.append(output_file)
+
+        # Run FFmpeg
+        try:
+            subprocess.run(ffmpeg_cmd, check=True)
+            print(f"Video created successfully: {output_file}")
+        except subprocess.CalledProcessCode as e:
+            print(f"Error creating video: {e}")
+        finally:
+            # Clean up temporary files
+            # for file in os.listdir(temp_dir):
+            #     os.remove(os.path.join(temp_dir, file))
+            # os.rmdir(temp_dir)
+            print("Temporary files not removed for debugging purposes.")
+
+        return (comment,)