"""The interface between scenes and ffmpeg.""" from __future__ import annotations __all__ = ["SceneFileWriter"] import json import shutil from fractions import Fraction from pathlib import Path from queue import Queue from tempfile import NamedTemporaryFile, _TemporaryFileWrapper from threading import Thread from typing import TYPE_CHECKING, Any import av import numpy as np import srt from PIL import Image from pydub import AudioSegment from manim import __version__ from .. import config, logger from .._config.logger_utils import set_file_logger from ..constants import RendererType from ..utils.file_ops import ( add_extension_if_not_present, add_version_before_extension, guarantee_existence, is_gif_format, is_png_format, modify_atime, write_to_movie, ) from ..utils.sounds import get_full_sound_file_path from .section import DefaultSectionType, Section if TYPE_CHECKING: from av.container.output import OutputContainer from av.stream import Stream from manim.renderer.cairo_renderer import CairoRenderer from manim.renderer.opengl_renderer import OpenGLRenderer from manim.typing import PixelArray, StrPath def to_av_frame_rate(fps: float) -> Fraction: epsilon1 = 1e-4 epsilon2 = 0.02 if isinstance(fps, int): (num, denom) = (fps, 1) elif abs(fps - round(fps)) < epsilon1: (num, denom) = (round(fps), 1) else: denom = 1001 num = round(fps * denom / 1000) * 1000 if abs(fps - num / denom) >= epsilon2: raise ValueError("invalid frame rate") return Fraction(num, denom) def convert_audio( input_path: Path, output_path: Path | _TemporaryFileWrapper[bytes], codec_name: str ) -> None: with ( av.open(input_path) as input_audio, av.open(output_path, "w") as output_audio, ): input_audio_stream = input_audio.streams.audio[0] output_audio_stream = output_audio.add_stream(codec_name) for frame in input_audio.decode(input_audio_stream): for packet in output_audio_stream.encode(frame): output_audio.mux(packet) for packet in output_audio_stream.encode(): output_audio.mux(packet) class SceneFileWriter: """SceneFileWriter is the object that actually writes the animations played, into video files, using FFMPEG. This is mostly for Manim's internal use. You will rarely, if ever, have to use the methods for this class, unless tinkering with the very fabric of Manim's reality. Attributes ---------- sections : list of :class:`.Section` used to segment scene sections_output_dir : :class:`pathlib.Path` where are section videos stored output_name : str name of movie without extension and basis for section video names Some useful attributes are: "write_to_movie" (bool=False) Whether or not to write the animations into a video file. "movie_file_extension" (str=".mp4") The file-type extension of the outputted video. "partial_movie_files" List of all the partial-movie files. """ force_output_as_scene_name = False def __init__( self, renderer: CairoRenderer | OpenGLRenderer, scene_name: str, **kwargs: Any, ) -> None: self.renderer = renderer self.init_output_directories(scene_name) self.init_audio() self.frame_count = 0 self.partial_movie_files: list[str | None] = [] self.subcaptions: list[srt.Subtitle] = [] self.sections: list[Section] = [] # first section gets automatically created for convenience # if you need the first section to be skipped, add a first section by hand, it will replace this one self.next_section( name="autocreated", type_=DefaultSectionType.NORMAL, skip_animations=False ) def init_output_directories(self, scene_name: str) -> None: """Initialise output directories. Notes ----- The directories are read from ``config``, for example ``config['media_dir']``. If the target directories don't already exist, they will be created. """ if config["dry_run"]: # in dry-run mode there is no output return module_name = config.get_dir("input_file").stem if config["input_file"] else "" if SceneFileWriter.force_output_as_scene_name: self.output_name = Path(scene_name) elif config["output_file"] and not config["write_all"]: self.output_name = config.get_dir("output_file") else: self.output_name = Path(scene_name) if config["media_dir"]: image_dir = guarantee_existence( config.get_dir( "images_dir", module_name=module_name, scene_name=scene_name ), ) self.image_file_path = image_dir / add_extension_if_not_present( self.output_name, ".png" ) if write_to_movie(): movie_dir = guarantee_existence( config.get_dir( "video_dir", module_name=module_name, scene_name=scene_name ), ) self.movie_file_path = movie_dir / add_extension_if_not_present( self.output_name, config["movie_file_extension"] ) # TODO: /dev/null would be good in case sections_output_dir is used without being set (doesn't work on Windows), everyone likes defensive programming, right? self.sections_output_dir = Path("") if config.save_sections: self.sections_output_dir = guarantee_existence( config.get_dir( "sections_dir", module_name=module_name, scene_name=scene_name ) ) if is_gif_format(): self.gif_file_path = add_extension_if_not_present( self.output_name, ".gif" ) if not config["output_file"]: self.gif_file_path = add_version_before_extension( self.gif_file_path ) self.gif_file_path = movie_dir / self.gif_file_path self.partial_movie_directory = guarantee_existence( config.get_dir( "partial_movie_dir", scene_name=scene_name, module_name=module_name, ), ) if config["log_to_file"]: log_dir = guarantee_existence(config.get_dir("log_dir")) set_file_logger( scene_name=scene_name, module_name=module_name, log_dir=log_dir ) def finish_last_section(self) -> None: """Delete current section if it is empty.""" if len(self.sections) and self.sections[-1].is_empty(): self.sections.pop() def next_section(self, name: str, type_: str, skip_animations: bool) -> None: """Create segmentation cut here.""" self.finish_last_section() # images don't support sections section_video: str | None = None # don't save when None if ( not config.dry_run and write_to_movie() and config.save_sections and not skip_animations ): # relative to index file section_video = f"{self.output_name}_{len(self.sections):04}_{name}{config.movie_file_extension}" self.sections.append( Section( type_, section_video, name, skip_animations, ), ) def add_partial_movie_file(self, hash_animation: str | None) -> None: """Adds a new partial movie file path to ``scene.partial_movie_files`` and current section from a hash. This method will compute the path from the hash. In addition to that it adds the new animation to the current section. Parameters ---------- hash_animation Hash of the animation. """ if not hasattr(self, "partial_movie_directory") or not write_to_movie(): return # None has to be added to partial_movie_files to keep the right index with scene.num_plays. # i.e if an animation is skipped, scene.num_plays is still incremented and we add an element to partial_movie_file be even with num_plays. if hash_animation is None: self.partial_movie_files.append(None) self.sections[-1].partial_movie_files.append(None) else: new_partial_movie_file = str( self.partial_movie_directory / f"{hash_animation}{config['movie_file_extension']}" ) self.partial_movie_files.append(new_partial_movie_file) self.sections[-1].partial_movie_files.append(new_partial_movie_file) def get_resolution_directory(self) -> str: """Get the name of the resolution directory directly containing the video file. This method gets the name of the directory that immediately contains the video file. This name is ``p``. For example, if you are rendering an 854x480 px animation at 15fps, the name of the directory that immediately contains the video, file will be ``480p15``. The file structure should look something like:: MEDIA_DIR |--Tex |--texts |--videos |-- |--p |--partial_movie_files |--.mp4 |--.srt Returns ------- :class:`str` The name of the directory. """ pixel_height = config["pixel_height"] frame_rate = config["frame_rate"] return f"{pixel_height}p{frame_rate}" # Sound def init_audio(self) -> None: """Preps the writer for adding audio to the movie.""" self.includes_sound = False def create_audio_segment(self) -> None: """Creates an empty, silent, Audio Segment.""" self.audio_segment = AudioSegment.silent() def add_audio_segment( self, new_segment: AudioSegment, time: float | None = None, gain_to_background: float | None = None, ) -> None: """This method adds an audio segment from an AudioSegment type object and suitable parameters. Parameters ---------- new_segment The audio segment to add time the timestamp at which the sound should be added. gain_to_background The gain of the segment from the background. """ if not self.includes_sound: self.includes_sound = True self.create_audio_segment() segment = self.audio_segment curr_end = segment.duration_seconds if time is None: time = curr_end if time < 0: raise ValueError("Adding sound at timestamp < 0") new_end = time + new_segment.duration_seconds diff = new_end - curr_end if diff > 0: segment = segment.append( AudioSegment.silent(int(np.ceil(diff * 1000))), crossfade=0, ) self.audio_segment = segment.overlay( new_segment, position=int(1000 * time), gain_during_overlay=gain_to_background, ) def add_sound( self, sound_file: StrPath, time: float | None = None, gain: float | None = None, **kwargs: Any, ) -> None: """This method adds an audio segment from a sound file. Parameters ---------- sound_file The path to the sound file. time The timestamp at which the audio should be added. gain The gain of the given audio segment. **kwargs This method uses add_audio_segment, so any keyword arguments used there can be referenced here. """ file_path = get_full_sound_file_path(sound_file) # we assume files with .wav / .raw suffix are actually # .wav and .raw files, respectively. if file_path.suffix not in (".wav", ".raw"): # we need to pass delete=False to work on Windows # TODO: figure out a way to cache the wav file generated (benchmark needed) with NamedTemporaryFile(suffix=".wav", delete=False) as wav_file_path: convert_audio(file_path, wav_file_path, "pcm_s16le") new_segment = AudioSegment.from_file(wav_file_path.name) logger.info(f"Automatically converted {file_path} to .wav") Path(wav_file_path.name).unlink() else: new_segment = AudioSegment.from_file(file_path) if gain: new_segment = new_segment.apply_gain(gain) self.add_audio_segment(new_segment, time, **kwargs) # Writers def begin_animation( self, allow_write: bool = False, file_path: StrPath | None = None ) -> None: """Used internally by manim to stream the animation to FFMPEG for displaying or writing to a file. Parameters ---------- allow_write Whether or not to write to a video file. """ if write_to_movie() and allow_write: self.open_partial_movie_stream(file_path=file_path) def end_animation(self, allow_write: bool = False) -> None: """Internally used by Manim to stop streaming to FFMPEG gracefully. Parameters ---------- allow_write Whether or not to write to a video file. """ if write_to_movie() and allow_write: self.close_partial_movie_stream() def listen_and_write(self) -> None: """For internal use only: blocks until new frame is available on the queue.""" while True: num_frames, frame_data = self.queue.get() if frame_data is None: break self.encode_and_write_frame(frame_data, num_frames) def encode_and_write_frame(self, frame: PixelArray, num_frames: int) -> None: """For internal use only: takes a given frame in ``np.ndarray`` format and writes it to the stream """ for _ in range(num_frames): # Notes: precomputing reusing packets does not work! # I.e., you cannot do `packets = encode(...)` # and reuse it, as it seems that `mux(...)` # consumes the packet. # The same issue applies for `av_frame`, # reusing it renders weird-looking frames. av_frame = av.VideoFrame.from_ndarray(frame, format="rgba") for packet in self.video_stream.encode(av_frame): self.video_container.mux(packet) def write_frame( self, frame_or_renderer: PixelArray | OpenGLRenderer, num_frames: int = 1 ) -> None: """Used internally by Manim to write a frame to the FFMPEG input buffer. Parameters ---------- frame_or_renderer Pixel array of the frame. num_frames The number of times to write frame. """ if write_to_movie(): if isinstance(frame_or_renderer, np.ndarray): frame = frame_or_renderer else: frame = ( frame_or_renderer.get_frame() if config.renderer == RendererType.OPENGL else frame_or_renderer ) msg = (num_frames, frame) self.queue.put(msg) if is_png_format() and not config["dry_run"]: if isinstance(frame_or_renderer, np.ndarray): image = Image.fromarray(frame_or_renderer) else: image = ( frame_or_renderer.get_image() if config.renderer == RendererType.OPENGL else Image.fromarray(frame_or_renderer) ) target_dir = self.image_file_path.parent / self.image_file_path.stem extension = self.image_file_path.suffix self.output_image( image, target_dir, extension, config["zero_pad"], ) def output_image( self, image: Image.Image, target_dir: StrPath, ext: str, zero_pad: int ) -> None: if zero_pad: image.save(f"{target_dir}{str(self.frame_count).zfill(zero_pad)}{ext}") else: image.save(f"{target_dir}{self.frame_count}{ext}") self.frame_count += 1 def save_image(self, image: Image.Image) -> None: """This method saves the image passed to it in the default image directory. Parameters ---------- image The pixel array of the image to save. """ if config["dry_run"]: return if not config["output_file"]: self.image_file_path = add_version_before_extension(self.image_file_path) image.save(self.image_file_path) self.print_file_ready_message(self.image_file_path) def finish(self) -> None: """Finishes writing to the FFMPEG buffer or writing images to output directory. Combines the partial movie files into the whole scene. If save_last_frame is True, saves the last frame in the default image directory. """ if write_to_movie(): self.combine_to_movie() if config.save_sections: self.combine_to_section_videos() if config["flush_cache"]: self.flush_cache_directory() else: self.clean_cache() elif is_png_format() and not config["dry_run"]: target_dir = self.image_file_path.parent / self.image_file_path.stem logger.info("\n%i images ready at %s\n", self.frame_count, str(target_dir)) if self.subcaptions: self.write_subcaption_file() def open_partial_movie_stream(self, file_path: StrPath | None = None) -> None: """Open a container holding a video stream. This is used internally by Manim initialize the container holding the video stream of a partial movie file. """ if file_path is None: file_path = self.partial_movie_files[self.renderer.num_plays] self.partial_movie_file_path = file_path fps = to_av_frame_rate(config.frame_rate) partial_movie_file_codec = "libx264" partial_movie_file_pix_fmt = "yuv420p" av_options = { "an": "1", # ffmpeg: -an, no audio "crf": "23", # ffmpeg: -crf, constant rate factor (improved bitrate) } if config.movie_file_extension == ".webm": partial_movie_file_codec = "libvpx-vp9" av_options["-auto-alt-ref"] = "1" if config.transparent: partial_movie_file_pix_fmt = "yuva420p" elif config.transparent: partial_movie_file_codec = "qtrle" partial_movie_file_pix_fmt = "argb" with av.open(file_path, mode="w") as video_container: stream = video_container.add_stream( partial_movie_file_codec, rate=fps, options=av_options, ) stream.pix_fmt = partial_movie_file_pix_fmt stream.width = config.pixel_width stream.height = config.pixel_height self.video_container: OutputContainer = video_container self.video_stream: Stream = stream self.queue: Queue[tuple[int, PixelArray | None]] = Queue() self.writer_thread = Thread(target=self.listen_and_write, args=()) self.writer_thread.start() def close_partial_movie_stream(self) -> None: """Close the currently opened video container. Used internally by Manim to first flush the remaining packages in the video stream holding a partial file, and then close the corresponding container. """ self.queue.put((-1, None)) self.writer_thread.join() for packet in self.video_stream.encode(): self.video_container.mux(packet) self.video_container.close() logger.info( f"Animation {self.renderer.num_plays} : Partial movie file written in %(path)s", {"path": f"'{self.partial_movie_file_path}'"}, ) def is_already_cached(self, hash_invocation: str) -> bool: """Will check if a file named with `hash_invocation` exists. Parameters ---------- hash_invocation The hash corresponding to an invocation to either `scene.play` or `scene.wait`. Returns ------- :class:`bool` Whether the file exists. """ if not hasattr(self, "partial_movie_directory") or not write_to_movie(): return False path = ( self.partial_movie_directory / f"{hash_invocation}{config['movie_file_extension']}" ) return path.exists() def combine_files( self, input_files: list[str], output_file: Path, create_gif: bool = False, includes_sound: bool = False, ) -> None: file_list = self.partial_movie_directory / "partial_movie_file_list.txt" logger.debug( f"Partial movie files to combine ({len(input_files)} files): %(p)s", {"p": input_files[:5]}, ) with file_list.open("w", encoding="utf-8") as fp: fp.write("# This file is used internally by FFMPEG.\n") for pf_path in input_files: pf_path = Path(pf_path).as_posix() fp.write(f"file 'file:{pf_path}'\n") av_options = { "safe": "0", # needed to read files } if not includes_sound: av_options["an"] = "1" partial_movies_input = av.open( str(file_list), options=av_options, format="concat" ) partial_movies_stream = partial_movies_input.streams.video[0] output_container = av.open(str(output_file), mode="w") output_container.metadata["comment"] = ( f"Rendered with Manim Community v{__version__}" ) output_stream = output_container.add_stream( codec_name="gif" if create_gif else None, template=partial_movies_stream if not create_gif else None, ) if config.transparent and config.movie_file_extension == ".webm": output_stream.pix_fmt = "yuva420p" if create_gif: """The following solution was largely inspired from this comment https://github.com/imageio/imageio/issues/995#issuecomment-1580533018, and the following code https://github.com/imageio/imageio/blob/65d79140018bb7c64c0692ea72cb4093e8d632a0/imageio/plugins/pyav.py#L927-L996. """ output_stream.pix_fmt = "rgb8" if config.transparent: output_stream.pix_fmt = "pal8" output_stream.width = config.pixel_width output_stream.height = config.pixel_height output_stream.rate = to_av_frame_rate(config.frame_rate) graph = av.filter.Graph() input_buffer = graph.add_buffer(template=partial_movies_stream) split = graph.add("split") palettegen = graph.add("palettegen", "stats_mode=diff") paletteuse = graph.add( "paletteuse", "dither=bayer:bayer_scale=5:diff_mode=rectangle" ) output_sink = graph.add("buffersink") input_buffer.link_to(split) split.link_to(palettegen, 0, 0) # 1st input of split -> input of palettegen split.link_to(paletteuse, 1, 0) # 2nd output of split -> 1st input palettegen.link_to(paletteuse, 0, 1) # output of palettegen -> 2nd input paletteuse.link_to(output_sink) graph.configure() for frame in partial_movies_input.decode(video=0): graph.push(frame) graph.push(None) # EOF: https://github.com/PyAV-Org/PyAV/issues/886. frames_written = 0 while True: try: frame = graph.pull() if output_stream.codec_context.time_base is not None: frame.time_base = output_stream.codec_context.time_base frame.pts = frames_written frames_written += 1 output_container.mux(output_stream.encode(frame)) except av.error.EOFError: break for packet in output_stream.encode(): output_container.mux(packet) else: for packet in partial_movies_input.demux(partial_movies_stream): # We need to skip the "flushing" packets that `demux` generates. if packet.dts is None: continue packet.dts = None # This seems to be needed, as dts from consecutive # files may not be monotically increasing, so we let libav compute it. # We need to assign the packet to the new stream. packet.stream = output_stream output_container.mux(packet) partial_movies_input.close() output_container.close() def combine_to_movie(self) -> None: """Used internally by Manim to combine the separate partial movie files that make up a Scene into a single video file for that Scene. """ partial_movie_files = [el for el in self.partial_movie_files if el is not None] # NOTE: Here we should do a check and raise an exception if partial # movie file is empty. We can't, as a lot of stuff (in particular, in # tests) use scene initialization, and this error would be raised as # it's just an empty scene initialized. # determine output path movie_file_path = self.movie_file_path if is_gif_format(): movie_file_path = self.gif_file_path if len(partial_movie_files) == 0: # Prevent calling concat on empty list logger.info("No animations are contained in this scene.") return logger.info("Combining to Movie file.") self.combine_files( partial_movie_files, movie_file_path, is_gif_format(), self.includes_sound, ) # handle sound if self.includes_sound and config.format != "gif": sound_file_path = movie_file_path.with_suffix(".wav") # Makes sure sound file length will match video file self.add_audio_segment(AudioSegment.silent(0)) self.audio_segment.export( sound_file_path, format="wav", bitrate="312k", ) # Audio added to a VP9 encoded (webm) video file needs # to be encoded as vorbis or opus. Directly exporting # self.audio_segment with such a codec works in principle, # but tries to call ffmpeg via its CLI -- which we want # to avoid. This is why we need to do the conversion # manually. if config.movie_file_extension == ".webm": ogg_sound_file_path = sound_file_path.with_suffix(".ogg") convert_audio(sound_file_path, ogg_sound_file_path, "libvorbis") sound_file_path = ogg_sound_file_path elif config.movie_file_extension == ".mp4": # Similarly, pyav may reject wav audio in an .mp4 file; # convert to AAC. aac_sound_file_path = sound_file_path.with_suffix(".aac") convert_audio(sound_file_path, aac_sound_file_path, "aac") sound_file_path = aac_sound_file_path temp_file_path = movie_file_path.with_name( f"{movie_file_path.stem}_temp{movie_file_path.suffix}" ) av_options = { "shortest": "1", "metadata": f"comment=Rendered with Manim Community v{__version__}", } with ( av.open(movie_file_path) as video_input, av.open(sound_file_path) as audio_input, ): video_stream = video_input.streams.video[0] audio_stream = audio_input.streams.audio[0] output_container = av.open( str(temp_file_path), mode="w", options=av_options ) output_video_stream = output_container.add_stream(template=video_stream) output_audio_stream = output_container.add_stream(template=audio_stream) for packet in video_input.demux(video_stream): # We need to skip the "flushing" packets that `demux` generates. if packet.dts is None: continue # We need to assign the packet to the new stream. packet.stream = output_video_stream output_container.mux(packet) for packet in audio_input.demux(audio_stream): # We need to skip the "flushing" packets that `demux` generates. if packet.dts is None: continue # We need to assign the packet to the new stream. packet.stream = output_audio_stream output_container.mux(packet) output_container.close() shutil.move(str(temp_file_path), str(movie_file_path)) sound_file_path.unlink() self.print_file_ready_message(str(movie_file_path)) if write_to_movie(): for file_path in partial_movie_files: # We have to modify the accessed time so if we have to clean the cache we remove the one used the longest. modify_atime(file_path) def combine_to_section_videos(self) -> None: """Concatenate partial movie files for each section.""" self.finish_last_section() sections_index: list[dict[str, Any]] = [] for section in self.sections: # only if section does want to be saved if section.video is not None: logger.info(f"Combining partial files for section '{section.name}'") self.combine_files( section.get_clean_partial_movie_files(), self.sections_output_dir / section.video, ) sections_index.append(section.get_dict(self.sections_output_dir)) with (self.sections_output_dir / f"{self.output_name}.json").open("w") as file: json.dump(sections_index, file, indent=4) def clean_cache(self) -> None: """Will clean the cache by removing the oldest partial_movie_files.""" cached_partial_movies = [ (self.partial_movie_directory / file_name) for file_name in self.partial_movie_directory.iterdir() if file_name != "partial_movie_file_list.txt" ] if len(cached_partial_movies) > config["max_files_cached"]: number_files_to_delete = ( len(cached_partial_movies) - config["max_files_cached"] ) oldest_files_to_delete = sorted( cached_partial_movies, key=lambda path: path.stat().st_atime, )[:number_files_to_delete] for file_to_delete in oldest_files_to_delete: file_to_delete.unlink() logger.info( f"The partial movie directory is full (> {config['max_files_cached']} files). Therefore, manim has removed the {number_files_to_delete} oldest file(s)." " You can change this behaviour by changing max_files_cached in config.", ) def flush_cache_directory(self) -> None: """Delete all the cached partial movie files""" cached_partial_movies = [ self.partial_movie_directory / file_name for file_name in self.partial_movie_directory.iterdir() if file_name != "partial_movie_file_list.txt" ] for f in cached_partial_movies: f.unlink() logger.info( f"Cache flushed. {len(cached_partial_movies)} file(s) deleted in %(par_dir)s.", {"par_dir": self.partial_movie_directory}, ) def write_subcaption_file(self) -> None: """Writes the subcaption file.""" if config.output_file is None: return subcaption_file = Path(config.output_file).with_suffix(".srt") subcaption_file.write_text(srt.compose(self.subcaptions), encoding="utf-8") logger.info(f"Subcaption file has been written as {subcaption_file}") def print_file_ready_message(self, file_path: StrPath) -> None: """Prints the "File Ready" message to STDOUT.""" config["output_file"] = file_path logger.info("\nFile ready at %(file_path)s\n", {"file_path": f"'{file_path}'"})