Recherche avancée

Sur d’autres sites (14112)

• Video generation and ffmpeg and locally stored images [closed]

  3 juillet, par Rahul Patil

  I am facing issue in ffmpeg while running the below code

  


  I'm working on a Flask application that generates a video by combining a sequence of images from a folder and a synthesized audio track using Bark (suno/bark-small). The idea is to use FFmpeg to stitch the images into a video, apply padding and scaling, and then merge it with the generated audio. I'm triggering the /generate-video endpoint with a simple curl POST request, passing a script that gets converted to audio. While the image and audio processing work as expected, FFmpeg fails during execution, and the server returns a 500 error. I've added error logging to capture FFmpeg’s stderr output, which suggests something is going wrong either with the generated input.txt file or the format of the inputs passed to FFmpeg. I'm not sure if the issue is related to file paths, the concat demuxer formatting, or possibly audio/video duration mismatch. Any insights on how to debug or correct the FFmpeg command would be appreciated.

  


  the curl request is

  


  curl -X POST http://localhost:5000/generate-video \
     -H "Content-Type: application/json" \
     -d '{"script": "Hello, this is a test script to generate a video."}' \
     --output output_video.mp4


  


  import os
import uuid
import subprocess
from pathlib import Path
import numpy as np
from flask import Flask, request, jsonify, send_file
from transformers import AutoProcessor, AutoModelForTextToWaveform
from scipy.io.wavfile import write as write_wav
import torch

# ========== CONFIG ==========
IMAGE_FOLDER = "./images"
OUTPUT_FOLDER = "./output"
RESOLUTION = (1280, 720)
IMAGE_DURATION = 3  # seconds per image
SAMPLE_RATE = 24000

app = Flask(__name__)
os.makedirs(OUTPUT_FOLDER, exist_ok=True)

# Load Bark-small model and processor
device = "cuda" if torch.cuda.is_available() else "cpu"
processor = AutoProcessor.from_pretrained("suno/bark-small")
model = AutoModelForTextToWaveform.from_pretrained("suno/bark-small").to(device)


# ========== UTILS ==========
def run_ffmpeg(cmd):
    result = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
    if result.returncode != 0:
        print("[FFmpeg ERROR]\n", result.stderr.decode())
        raise RuntimeError("FFmpeg failed.")
    else:
        print("[FFmpeg] Success.")


def find_images(folder):
    return sorted([
        f for f in Path(folder).glob("*")
        if f.suffix.lower() in {".jpg", ".jpeg", ".png"}
    ])


def create_ffmpeg_input_list(images, list_file_path):
    with open(list_file_path, "w") as f:
        for img in images:
            f.write(f"file '{img.resolve()}'\n")
            f.write(f"duration {IMAGE_DURATION}\n")
        # Repeat last image to avoid cutoff
        f.write(f"file '{images[-1].resolve()}'\n")


# ========== FLASK ROUTE ==========
@app.route('/generate-video', methods=['POST'])
def generate_video():
    data = request.get_json()
    script = data.get("script")
    if not script:
        return jsonify({"error": "No script provided"}), 400

    images = find_images(IMAGE_FOLDER)
    if not images:
        return jsonify({"error": "No images found in ./images"}), 400

    # Generate audio
    print("[1/3] Generating audio with Bark...")
    inputs = processor(script, return_tensors="pt").to(device)
    with torch.no_grad():
        audio_values = model.generate(**inputs)

    audio_np = audio_values[0].cpu().numpy().squeeze()
    audio_np = np.clip(audio_np, -1.0, 1.0)
    audio_int16 = (audio_np * 32767).astype(np.int16)

    audio_path = os.path.join(OUTPUT_FOLDER, f"{uuid.uuid4()}.wav")
    write_wav(audio_path, SAMPLE_RATE, audio_int16)

    # Create FFmpeg concat file
    print("[2/3] Preparing image list for FFmpeg...")
    list_file = os.path.join(OUTPUT_FOLDER, "input.txt")
    create_ffmpeg_input_list(images, list_file)

    # Final video path
    final_video_path = os.path.join(OUTPUT_FOLDER, f"{uuid.uuid4()}.mp4")

    # Run FFmpeg
    print("[3/3] Running FFmpeg to create video...")
    ffmpeg_cmd = [
        "ffmpeg", "-y",
        "-f", "concat", "-safe", "0", "-i", list_file,
        "-i", audio_path,
        "-vf", f"scale={RESOLUTION[0]}:{RESOLUTION[1]}:force_original_aspect_ratio=decrease,"
               f"pad={RESOLUTION[0]}:{RESOLUTION[1]}:(ow-iw)/2:(oh-ih)/2:color=black",
        "-c:v", "libx264", "-pix_fmt", "yuv420p",
        "-c:a", "aac", "-b:a", "192k",
        "-shortest", "-movflags", "+faststart",
        final_video_path
    ]

    try:
        run_ffmpeg(ffmpeg_cmd)
    except RuntimeError:
        return jsonify({"error": "FFmpeg failed. Check server logs."}), 500

    return send_file(final_video_path, as_attachment=True)


# ========== RUN APP ==========
if __name__ == '__main__':
    app.run(debug=True)


  


• Evolution #3822 : ventiler les CSS concernant les formulaires de spip.css -> form.css

  14 août 2017, par tetue tetue

  Voir aussi : #3964

• cbs : Describe allocate/free methods in tabular form

  27 juillet 2020, par Mark Thompson

  cbs : Describe allocate/free methods in tabular form
  Unit types are split into three categories, depending on how their
  
  content is managed :
  
  * POD structure - these require no special treatment.
  
  * Structure containing references to refcounted buffers - these can use
  
    a common free function when the offsets of all the internal references
  
    are known.
  
  * More complex structures - these still require ad-hoc treatment.
  For each codec we can then maintain a table of descriptors for each set of
  
  equivalent unit types, defining the mechanism needed to allocate/free that
  
  unit content.  This is not required to be used immediately - a new alloc
  
  function supports this, but does not replace the old one which works without
  
  referring to these tables.
  

  • [DH] libavcodec/cbs.c
  • [DH] libavcodec/cbs.h
  • [DH] libavcodec/cbs_internal.h