How to use Blob URL, MediaSource or other methods to play concatenated Blobs of media fragments?

There is currently no Web API targeted to video editing.
The MediaStream and MediaRecorder APIs are meant to deal with live sources.

Because of the structure of video files, you can't just slice a part of it to make a new video, nor can you just concatenate small video files to make one longer. In both cases, you need to rebuild its metadata in order to make a new video file.
The only current API able to produce MediaFiles is the MediaRecorder.

There is currently only two implementors of the MediaRecorder API, but they support about 3 different codecs in two different containers, which does mean that you would need to build yourself at least 5 metadata parsers to only support current implementations (which will keep growing in number, and which may need update as implementations are updated).
Sounds like a tough job.

Maybe the incoming WebAssembly API will allow us to port ffmpeg to browsers, which would make it a lot simpler, but I have to admit I don't know WA at all, so I'm not even sure it is really doable.


I hear you saying "Ok, there is no tool made just for that, but we are hackers, and we have other tools, with great power."
Well, yes. If we're really willing to do it, we can hack something...

As said before, the MediaStream and MediaRecorder are meant for live video. We can thus convert static video files to live streams with the [HTMLVideoElement | HTMLCanvasElement].captureStream() methods.
We can also record those live-streams to a static File thanks to the MediaRecorder API.

What we cannot do however is to change the current stream-source a MediaRecorder as been fed with.

So in order to merge small video Files into one longer, we'll need to

  • load these videos into <video> elements
  • draw these <video> elements on a <canvas> element in wanted order
  • feed an AudioContext's stream source with the <video> elements
  • merge the canvas.captureStream and AudioStreamSource's streams in a single MediaStream
  • Record this MediaStream

But this means that the merging is actually a re-recording of all the videos, and this can only be done in real-time (speed = x1)


Here is a live proof of concept where we first slice an original Video File in multiple smaller parts, shuffle these parts to mimic some montage, then create a canvas based player, also able to record this montage and export it.

NotaBene : This is the first version, and I still have a lot of bugs (notabely in Firefox, should work almost fine in chrome).

(() => {
  if (!('MediaRecorder' in window)) {
    throw new Error('unsupported browser');
  }
  // some global params
  const CHUNK_DURATION = 1000;
  const MAX_SLICES = 15; // get only 15 slices
  const FPS = 30;

  async function init() {
    const url = 'https://nickdesaulniers.github.io/netfix/demo/frag_bunny.mp4';
    const slices = await getSlices(url); // slice the original media in longer chunks
    mess_up_array(slices); // Let's shuffle these slices,
    // otherwise there is no point merging it in a new file
    generateSelect(slices); // displays each chunk independentely
    window.player = new SlicePlayer(slices); // init our player
  };

  const SlicePlayer = class {
    /*
   		@args: Array of populated HTMLVideoElements
	*/
    constructor(parts) {
      this.parts = parts;

      this.initVideoContext();
      this.initAudioContext();

      this.currentIndex = 0; // to know which video we'll play
      this.currentTime = 0;

      this.duration = parts.reduce((a, b) => b._duration + a, 0); // the sum of all parts' durations
      // (see below why "_")

      this.initDOM();
      // attach our onended callback only on the last vid
      this.parts[this.parts.length - 1].onended = e => this.onended();
      this.resetAll(); // set all videos' currentTime to 0 + draw first frame
    }
    initVideoContext() {
      const c = this.canvas = document.createElement('canvas');
      c.width = this.parts[0].videoWidth;
      c.height = this.parts[0].videoHeight;
      this.v_ctx = c.getContext('2d');
    }
    initAudioContext() {
      const a = this.a_ctx = new AudioContext();
      const gain = this.volume_node = a.createGain();
      gain.connect(a.destination);
      // extract the audio from our video elements so that we can record it
      this.audioSources = this.parts.map(v => a.createMediaElementSource(v));
      this.audioSources.forEach(s => s.connect(gain));
    }
    initDOM() {
      // all DOM things...
      canvas_player_timeline.max = this.duration;
      canvas_player_cont.appendChild(this.canvas);
      canvas_player_play_btn.onclick = e => this.startVid(this.currentIndex);
      canvas_player_cont.style.display = 'inline-block';
      canvas_player_timeline.oninput = e => {
        if (!this.recording)
          this.onseeking(e);
      };
      canvas_player_record_btn.onclick = e => this.record();
    }
    resetAll() {
      this.currentTime = canvas_player_timeline.value = 0;
      // when the first part as actually been reset to start
      this.parts[0].onseeked = e => {
        this.parts[0].onseeked = null;
        this.draw(0); // draw it
      };
      this.parts.forEach(v => v.currentTime = 0);

      if (this.playing && this.stopLoop) {
        this.playing = false;
        this.stopLoop();
      }
    }
    startVid(index) { // starts playing the video at given index
      if (index > this.parts.length - 1) { // that was the last one
        this.onended();
        return;
      }
      this.playing = true;
      this.currentIndex = index; // update our currentIndex
      this.parts[index].play().then(() => {
        // try to avoid at maximum the gaps between different parts
        if (this.recording && this.recorder.state === 'paused') {
          this.recorder.resume();
        }
      });

      this.startLoop();
    }
    startNext() { // starts the next part before the current one actually ended
      const nextPart = this.parts[this.currentIndex + 1];
      if (!nextPart) { // current === last
        return;
      }
      this.playing = true;
      if (!nextPart.paused) { // already playing ?
        return;
      }
      // try to avoid at maximum the gaps between different parts
      if (this.recording && this.recorder && this.recorder.state === 'recording') {
        this.recorder.pause();
      }

      nextPart.play()
        .then(() => {
          ++this.currentIndex; // this is now the current video
          if (!this.playing) { // somehow got stop in between ?
            this.playing = true;
            this.startLoop(); // start again
          }
          // try to avoid at maximum the gaps between different parts
          if (this.recording && this.recorder.state === 'paused') {
            this.recorder.resume();
          }
        });

    }
    startLoop() { // starts our update loop
      // see https://stackoverflow.com/questions/40687010/
      this.stopLoop = audioTimerLoop(e => this.update(), 1000 / FPS);
    }
    update(t) { // at every tick
      const currentPart = this.parts[this.currentIndex];

      this.updateTimeLine(); // update the timeline

      if (!this.playing || currentPart.paused) { // somehow got stopped
        this.playing = false;
        if (this.stopLoop) {
          this.stopLoop(); // stop the loop
        }
      }

      this.draw(this.currentIndex); // draw the current video on the canvas

      // calculate how long we've got until the end of this part
      const remainingTime = currentPart._duration - currentPart.currentTime;
      if (remainingTime < (2 / FPS)) { // less than 2 frames ?
        setTimeout(e => this.startNext(), remainingTime / 2); // start the next part
      }
    }
    draw(index) { // draw the video[index] on the canvas
      this.v_ctx.drawImage(this.parts[index], 0, 0);
    }
    updateTimeLine() {
      // get the sum of all parts' currentTime
      this.currentTime = this.parts.reduce((a, b) =>
        (isFinite(b.currentTime) ? b.currentTime : b._duration) + a, 0);
      canvas_player_timeline.value = this.currentTime;
    }
    onended() { // triggered when the last part ends
      // if we are recording, stop the recorder
      if (this.recording && this.recorder.state !== 'inactive') {
        this.recorder.stop();
      }
      // go back to first frame
      this.resetAll();
      this.currentIndex = 0;
      this.playing = false;
    }
    onseeking(evt) { // when we click the timeline
      // first reset all videos' currentTime to 0
      this.parts.forEach(v => v.currentTime = 0);
      this.currentTime = +evt.target.value;
      let index = 0;
      let sum = 0;
      // find which part should be played at this time
      for (index; index < this.parts.length; index++) {
        let p = this.parts[index];
        if (sum + p._duration > this.currentTime) {
          break;
        }
        sum += p._duration;
        p.currentTime = p._duration;
      }
      this.currentIndex = index;
      // set the currentTime of this part
      this.parts[index].currentTime = this.currentTime - sum;

      if (this.playing) { // if we were playing
        this.startVid(index); // set this part as the current one
      } else {
        this.parts[index].onseeked = e => { // wait we actually seeked the correct position
          this.parts[index].onseeked = null;
          this.draw(index); // and draw a single frame
        };
      }
    }
    record() { // inits the recording
      this.recording = true; // let the app know we're recording
      this.resetAll(); // go back to first frame

      canvas_controls.classList.add('disabled'); // disable controls

      const v_stream = this.canvas.captureStream(FPS); // make a stream of our canvas
      const dest = this.a_ctx.createMediaStreamDestination(); // make a stream of our AudioContext
      this.volume_node.connect(dest);
      // FF bug... see https://bugzilla.mozilla.org/show_bug.cgi?id=1296531
      let merged_stream = null;
      if (!('mozCaptureStream' in HTMLVideoElement.prototype)) {
        v_stream.addTrack(dest.stream.getAudioTracks()[0]);
        merged_stream = v_stream;
      } else {
        merged_stream = new MediaStream(
          v_stream.getVideoTracks().concat(dest.stream.getAudioTracks())
        );
      }

      const chunks = [];
      const rec = this.recorder = new MediaRecorder(merged_stream, {
        mimeType: MediaRecorder._preferred_type
      });
      rec.ondataavailable = e => chunks.push(e.data);
      rec.onstop = e => {
        merged_stream.getTracks().forEach(track => track.stop());
        this.export(new Blob(chunks));
      }
      rec.start();

      this.startVid(0); // start playing
    }
    export (blob) { // once the recording is over
      const a = document.createElement('a');
      a.download = a.innerHTML = 'merged.webm';
      a.href = URL.createObjectURL(blob, {
        type: MediaRecorder._preferred_type
      });
      exports_cont.appendChild(a);
      canvas_controls.classList.remove('disabled');
      this.recording = false;
      this.resetAll();
    }
  }

  // END Player

  function generateSelect(slices) { // generates a select to show each slice independently
    const select = document.createElement('select');
    select.appendChild(new Option('none', -1));
    slices.forEach((v, i) => select.appendChild(new Option(`slice ${i}`, i)));
    document.body.insertBefore(select, slice_player_cont);
    select.onchange = e => {
      slice_player_cont.firstElementChild && slice_player_cont.firstElementChild.remove();
      if (+select.value === -1) return; // 'none'
      slice_player_cont.appendChild(slices[+select.value]);
    };
  }

  async function getSlices(url) { // loads the main video, and record some slices from it

    const mainVid = await loadVid(url);

    // try to make the slicing silent... That's not easy.
    let a = null;
    if (mainVid.mozCaptureStream) { // target FF
      a = new AudioContext();
      // this causes an Range error in chrome
      //		a.createMediaElementSource(mainVid);
    } else { // chrome
      // this causes the stream to be muted too in FF
      mainVid.muted = true;
      // mainVid.volume = 0; // same
    }

    mainVid.play();
    const mainStream = mainVid.captureStream ? mainVid.captureStream() : mainVid.mozCaptureStream();
    console.log('mainVid loaded');
    const slices = await getSlicesInLoop(mainStream, mainVid);
    console.log('all slices loaded');
    setTimeout(() => console.clear(), 1000);
    if (a && a.close) { // kill the silence audio context (FF)
      a.close();
    }
    mainVid.pause();
    URL.revokeObjectURL(mainVid.src);

    return Promise.resolve(slices);
  }

  async function getSlicesInLoop(stream, mainVid) { // far from being precise
    // to do it well, we would need to get the keyframes info, but it's out of scope for this answer
    let slices = [];
    const loop = async function(i) {
      const slice = await mainVid.play().then(() => getNewSlice(stream, mainVid));
      console.log(`${i + 1} slice(s) loaded`);
      slices.push(slice);
      if ((mainVid.currentTime < mainVid._duration) && (i + 1 < MAX_SLICES)) {
        loop(++i);
      } else done(slices);
    };
    loop(0);
    let done;
    return new Promise((res, rej) => {
      done = arr => res(arr);
    });
  }

  function getNewSlice(stream, vid) { // one recorder per slice
    return new Promise((res, rej) => {
      const rec = new MediaRecorder(stream, {
        mimeType: MediaRecorder._preferred_type
      });
      const chunks = [];
      rec.ondataavailable = e => chunks.push(e.data);
      rec.onstop = e => {
        const blob = new Blob(chunks);
        res(loadVid(URL.createObjectURL(blob)));
      }
      rec.start();
      setTimeout(() => {
        const p = vid.pause();
        if (p && p.then)
          p.then(() => rec.stop())
        else
          rec.stop()
      }, CHUNK_DURATION);
    });
  }

  function loadVid(url) { // helper returning an video, preloaded
    return fetch(url)
      .then(r => r.blob())
      .then(b => makeVid(URL.createObjectURL(b)))
  };

  function makeVid(url) { // helper to create a video element
    const v = document.createElement('video');
    v.control = true;
    v.preload = 'metadata';
    return new Promise((res, rej) => {
      v.onloadedmetadata = e => {
        // chrome duration bug...
        // see https://bugs.chromium.org/p/chromium/issues/detail?id=642012
        // will also occur in next FF versions, in worse...
        if (v.duration === Infinity) {
          v.onseeked = e => {
            v._duration = v.currentTime; // FF new bug never updates duration to correct value
            v.onseeked = null;
            v.currentTime = 0;
            res(v);
          };
          v.currentTime = 1e5; // big but not too big either
        } else {
          v._duration = v.duration;
          res(v);
        }
      };
      v.onerror = rej;
      v.src = url;
    });
  };

  function mess_up_array(arr) { // shuffles an array
	const _sort = () => {
      let r = Math.random() - .5;
      return r < -0.1 ? -1 : r > 0.1 ? 1 : 0;
    };
    arr.sort(_sort)
    arr.sort(_sort)
    arr.sort(_sort);
  }

  /*
      An alternative timing loop, based on AudioContext's clock

      @arg callback : a callback function 
          with the audioContext's currentTime passed as unique argument
      @arg frequency : float in ms;
      @returns : a stop function

  */
  function audioTimerLoop(callback, frequency) {

    const freq = frequency / 1000; // AudioContext time parameters are in seconds
    const aCtx = new AudioContext();
    // Chrome needs our oscillator node to be attached to the destination
    // So we create a silent Gain Node
    const silence = aCtx.createGain();
    silence.gain.value = 0;
    silence.connect(aCtx.destination);

    onOSCend();

    var stopped = false; // A flag to know when we'll stop the loop
    function onOSCend() {
      const osc = aCtx.createOscillator();
      osc.onended = onOSCend; // so we can loop
      osc.connect(silence);
      osc.start(0); // start it now
      osc.stop(aCtx.currentTime + freq); // stop it next frame
      callback(aCtx.currentTime); // one frame is done
      if (stopped) { // user broke the loop
        osc.onended = function() {
          aCtx.close(); // clear the audioContext
          return;
        };
      }
    };
    // return a function to stop our loop
    return () => stopped = true;
  }

  // get the preferred codec available (vp8 is my personal, more reader support)
  MediaRecorder._preferred_type = [
      "video/webm\;codecs=vp8",
      "video/webm\;codecs=vp9",
      "video/webm\;codecs=h264",
      "video/webm"
    ]
    .filter(t => MediaRecorder.isTypeSupported(t))[0];


  init();

})();
#canvas_player_cont {
  display: none;
  position: relative;
}

#canvas_player_cont.disabled {
  opacity: .7;
  pointer-events: none;
}

#canvas_controls {
  position: absolute;
  bottom: 4px;
  left: 0px;
  width: calc(100% - 8px);
  display: flex;
  background: rgba(0, 0, 0, .7);
  padding: 4px;
}

#canvas_player_play_btn {
  flex-grow: 0;
}

#canvas_player_timeline {
  flex-grow: 1;
}
<div id="slice_player_cont">
</div>
<div id="canvas_player_cont">
  <div id="canvas_controls">
    <button id="canvas_player_play_btn">play</button>
    <input type="range" min="0" max="10" step="0.01" id="canvas_player_timeline">
    <button id="canvas_player_record_btn">save</button>
  </div>
</div>
<div id="exports_cont"></div>