+import io
from os import unlink
from tempfile import NamedTemporaryFile
from django.db import models
from django.utils.translation import gettext_lazy as _
from django.template import Template, Context
+import requests
from apiclient import youtube_call
from .utils import (
- video_from_image,
- cut_video,
+ concat_audio,
concat_videos,
+ cut_video,
get_duration,
get_framerate,
mux,
+ standardize_video,
+ video_from_image,
)
+from .thumbnail import create_thumbnail
class YouTube(models.Model):
title_template = models.CharField(max_length=1024, blank=True)
description_template = models.TextField(blank=True)
category = models.IntegerField(null=True, blank=True) # get categories
- intro_card = models.FileField(blank=True)
- intro_card_duration = models.FloatField(null=True, blank=True)
- card = models.FileField(blank=True)
- loop_video = models.FileField(blank=True)
- outro_card = models.FileField(blank=True)
- outro_card_duration = models.FloatField(null=True, blank=True)
+ intro_flac = models.FileField(upload_to='youtube/intro_flac', blank=True)
+ outro_flac = models.FileField(upload_to='youtube/outro_flac', blank=True)
+ loop_card = models.FileField(upload_to='youtube/card', blank=True)
+ loop_video = models.FileField(upload_to='youtube/loop_video', blank=True)
+ thumbnail_template = models.FileField(upload_to='youtube/thumbnail', blank=True)
+ thumbnail_definition = models.TextField(blank=True)
+ genres = models.CharField(max_length=2048, blank=True)
class Meta:
verbose_name = _("YouTube configuration")
"POST",
"https://www.googleapis.com/upload/youtube/v3/videos",
params={'part': part},
- data=data,
- media_data=f.read(),
+ json=data,
+ resumable_data=f.read(),
)
data = response.json()
audiobook.youtube_id = data['id']
audiobook.save(update_fields=['youtube_id'])
+
+ self.update_thumbnail(audiobook)
return response
def prepare_file(self, input_path, output_path=None):
- duration = get_duration(input_path)
+ audio = self.prepare_audio(input_path)
+ duration = self.get_duration(input_path)
video = self.prepare_video(duration)
- output = mux([video, input_path], output_path=output_path)
+ output = mux([video, audio], output_path=output_path)
+ unlink(audio)
unlink(video)
return output
+ def get_duration(self, input_path):
+ d = get_duration(input_path)
+ if self.intro_flac:
+ d += get_duration(self.intro_flac.path)
+ if self.outro_flac:
+ d += get_duration(self.outro_flac.path)
+ return d
+
+ def prepare_audio(self, input_path):
+ files = []
+ if self.intro_flac:
+ files.append(self.intro_flac.path)
+ files.append(input_path)
+ if self.outro_flac:
+ files.append(self.outro_flac.path)
+ return concat_audio(files)
+
def prepare_video(self, duration):
concat = []
+ outro = []
delete = []
if self.loop_video:
fps = get_framerate(self.loop_video.path)
+ loop_video = standardize_video(self.loop_video.path)
else:
fps = 25
loop_duration = duration
- if self.intro_card and self.intro_card_duration:
- loop_duration -= self.intro_card_duration
- intro = video_from_image(
- self.intro_card.path, self.intro_card_duration, fps=fps
+ for card in self.card_set.filter(duration__gt=0):
+ loop_duration -= card.duration
+ card_video = video_from_image(
+ card.image.path, card.duration, fps=fps
)
- concat.append(intro)
- delete.append(intro)
-
- if self.outro_card and self.outro_card_duration:
- loop_duration -= self.outro_card_duration
- outro = video_from_image(
- self.outro_card.path, self.outro_card_duration, fps=fps
- )
- concat.append(outro)
- delete.append(outro)
+ (concat if card.order < 0 else outro).append(card_video)
+ delete.append(card_video)
if self.loop_video:
- loop_video_duration = get_duration(self.loop_video.path)
+ loop_video_duration = get_duration(loop_video)
times_loop = int(loop_duration // loop_video_duration)
leftover_duration = loop_duration % loop_video_duration
- leftover = cut_video(self.loop_video.path, leftover_duration)
- concat[1:1] = [self.loop_video.path] * times_loop + [leftover]
+ leftover = cut_video(loop_video, leftover_duration)
+ concat.extend([loop_video] * times_loop + [leftover])
delete.append(leftover)
else:
- leftover = video_from_image(self.card.path, loop_duration)
- concat.insert(1, video_from_image(self.card.path, loop_duration, fps=fps))
+ leftover = video_from_image(self.loop_card.path, loop_duration)
+ concat.append(video_from_image(self.loop_card.path, loop_duration, fps=fps))
delete.append(leftover)
+ concat.extend(outro)
output = concat_videos(concat)
for p in delete:
unlink(p)
+ unlink(loop_video)
return output
# tags
# license
# selfDeclaredMadeForKids
+
+ def update_thumbnail(self, audiobook):
+ thumbnail = self.prepare_thumbnail(audiobook)
+ response = youtube_call(
+ "POST",
+ "https://www.googleapis.com/upload/youtube/v3/thumbnails/set",
+ params={'videoId': audiobook.youtube_id},
+ data=thumbnail.getvalue(),
+ )
+
+ def prepare_thumbnail(self, audiobook):
+ slug = audiobook.url.rstrip('/').rsplit('/', 1)[-1]
+ apidata = requests.get(f'https://wolnelektury.pl/api/books/{slug}/').json()
+ img = create_thumbnail(
+ self.thumbnail_template.path,
+ self.thumbnail_definition,
+ {
+ "author": ', '.join((a['name'] for a in apidata['authors'])),
+ "title": apidata['title'],
+ },
+ lambda name: Font.objects.get(name=name).truetype.path
+ )
+ buf = io.BytesIO()
+ img.save(buf, format='PNG')
+ return buf
+
+
+class Card(models.Model):
+ youtube = models.ForeignKey(YouTube, models.CASCADE)
+ order = models.SmallIntegerField()
+ image = models.FileField(upload_to='youtube/card')
+ duration = models.FloatField()
+
+ class Meta:
+ ordering = ('order', )
+
+
+class Font(models.Model):
+ name = models.CharField(max_length=255, unique=True)
+ truetype = models.FileField(upload_to='youtube/font')
+
+ def __str__(self):
+ return self.name