Files
shorts-autopilot/app.py
T

1790 lines
68 KiB
Python

import base64
import json
import logging
import os
import re
import shutil
import subprocess
import sys
import tempfile
import threading
from collections import deque
from datetime import date, datetime, timedelta, timezone
from logging.handlers import RotatingFileHandler
from pathlib import Path
import requests
from fastapi import FastAPI, Request
from fastapi.responses import FileResponse, JSONResponse, RedirectResponse
from google.auth.transport.requests import Request as GoogleRequest
from google.oauth2.credentials import Credentials
from google_auth_oauthlib.flow import Flow
from googleapiclient.discovery import build
from googleapiclient.http import MediaFileUpload
os.environ.setdefault("OAUTHLIB_INSECURE_TRANSPORT", "1")
os.environ.setdefault("OAUTHLIB_RELAX_TOKEN_SCOPE", "1")
BASE = Path(__file__).resolve().parent
DATA = BASE / "data"
LOGS = DATA / "logs"
PROFILES_DIR = DATA / "profiles"
for d in (DATA, LOGS, PROFILES_DIR):
d.mkdir(parents=True, exist_ok=True)
SETTINGS_FILE = DATA / "settings.json"
STATE_FILE = DATA / "state.json"
CLIENT_SECRET = BASE / "client_secret.json"
LOG_FILE = LOGS / "app.log"
REDIRECT_URI = os.environ.get("REDIRECT_URI", "http://localhost:8000/auth/callback")
SCOPES = [
"https://www.googleapis.com/auth/youtube.upload",
"https://www.googleapis.com/auth/youtube.readonly",
]
MASK = "********"
MIN_FREE_GB = 2
LATIN_LANGS = {"en", "es", "fr", "de", "it", "pt", "nl", "sv", "no", "da", "fi", "pl", "cs", "ro", "hu", "tr", "id", "ms", "vi", "tl"}
REVIEW_MODES = ("advisory", "block", "off")
ENGINES = ("local", "veo")
TTS_ENGINES = ("auto", "kokoro", "piper", "say", "espeak")
SUGGESTED_TEXT_MODELS = ["llama3.1:8b", "llama3.2:3b", "qwen2.5:7b", "mistral:7b"]
SUGGESTED_VISION_MODELS = ["moondream", "llava:7b", "llama3.2-vision", "qwen2.5vl:7b"]
SMALL_VISION_MODELS = ("moondream", "llava-phi3", "bakllava")
GLOBAL_DEFAULTS = {
"active_profile": "default",
"storage_dir": "videos",
"models_dir": "",
"temp_dir": "",
"delete_after_publish": False,
"keep_clips": False,
"video_engine": "local",
"ollama_url": "http://localhost:11434",
"ollama_model": "llama3.1:8b",
"vision_model": "moondream",
"ollama_timeout": 300,
"sd_model": "stabilityai/sdxl-turbo",
"sd_steps": 4,
"sd_guidance": 2,
"sd_width": 768,
"sd_height": 1344,
"tts_engine": "auto",
"tts_voice": "af_heart",
"tts_rate": 170,
"piper_model": "",
"auto_install": True,
"gemini_api_key": "",
"veo_model": "veo-3.1-fast-generate-preview",
"aspect_ratio": "9:16",
"target_seconds": 60,
"clip_seconds": 8,
"error_cooldown_minutes": 5,
}
PROFILE_DEFAULTS = {
"name": "Default",
"channel_niche": "",
"made_for_kids": False,
"kids_manual_review": True,
"kids_llm_review": "advisory",
"upload_category_id": "24",
"privacy_status": "private",
"trending_query": "",
"trending_days": 7,
"category_id": "",
"region_code": "US",
"source_language": "en",
"trending_count": 25,
"videos_per_day": 0,
"minutes_between_videos": 0,
}
KIDS_RULES = """The audience is young children under 13. Follow every one of these rules:
Use simple, warm, positive language a 5 year old understands.
Characters are cartoon animals, creatures, or objects. Never realistic humans and never real children.
No violence, weapons, injuries, blood, death, dangerous acts children could imitate, scary or disturbing imagery, bullying, romance, kissing, gross-out humor, alcohol, tobacco, drugs, or adult themes.
No brand names, real products, toys for sale, or product placement.
Never ask viewers to comment, like, subscribe, share personal information, visit a website, or leave YouTube.
No clickbait. The title honestly describes the story, uses no ALL CAPS words, and has at most one exclamation mark.
Include a gentle lesson or positive message such as kindness, sharing, curiosity, or courage.
Visuals are bright, colorful, calm, friendly animation in safe, cheerful settings, with no flashing or strobing light.
"""
REVIEW_RULES = {
"violence": "violence, weapons, injuries, blood, death, or dangerous acts a child could copy",
"scary": "scary, creepy, or disturbing imagery or events",
"humans": "realistic humans or real children as characters",
"adult": "romance, kissing, alcohol, tobacco, drugs, or other adult themes",
"commercial": "brand names, real products, toys for sale, or product placement",
"contact": "asking viewers to comment, like, subscribe, share personal information, visit a website, or leave YouTube",
"negative": "bullying, meanness, or a message that teaches bad behavior",
}
KIDS_VEO_SUFFIX = (
" Flat 2D cartoon animation, never photorealistic, correct anatomy with the right number of limbs. Child-friendly cartoon animation with cute non-human characters, soft bright colors, calm pacing, "
"safe cheerful setting, nothing scary or dangerous, no realistic people, no logos, no flashing lights."
)
CARTOON_SUFFIX = (
", flat 2D cartoon illustration, hand drawn animation style, simple stylized cartoon characters, "
"correct anatomy with the right number of limbs, one head per character, clean clear shapes, "
"not photorealistic, not a photograph, no realistic people"
)
KIDS_IMAGE_SUFFIX = ", children's picture book illustration, cute cartoon characters, bright soft colors, cheerful and safe, gentle"
IMAGE_NEGATIVE = (
"photorealistic, photograph, photo, realistic human, real person, lifelike face, hyperrealistic, 3d render, "
"extra limbs, extra legs, extra arms, extra heads, two heads, multiple heads, duplicated body, duplicate character, "
"fused limbs, missing limbs, malformed limbs, deformed, mutated, disfigured, bad anatomy, wrong anatomy, "
"malformed hands, extra fingers, distorted face, blurry, low quality, text, letters, watermark, logo, signature, "
"scary, creepy, horror, violence, weapon, blood, gore, nsfw"
)
KIDS_VISION_PROMPT = """Look carefully at this image from a video for young children.
First describe what you actually see in the image.
Then decide: it is unsafe if it shows photorealistic or lifelike people, realistic children, any character with the wrong number of limbs, legs, arms, or heads, duplicated or merged bodies, deformed or distorted faces, violence, weapons, blood, injury, scary or creepy imagery, nudity, alcohol, tobacco, drugs, brand logos, or anything a parent would find inappropriate for a 4 year old.
Count the legs and heads on each animal or person before deciding.
Reply with JSON only, filling in your own words: {"description": "what you see", "safe": true or false, "reason": "why"}"""
VISION_TEXT_PROMPT = """Look at this image, which is meant for young children.
Answer with one word, safe or unsafe, then a short reason.
Answer unsafe if it shows photorealistic or lifelike people, any character with the wrong number of limbs or heads, duplicated or merged bodies, violence, weapons, blood, injury, scary or creepy imagery, distorted faces, nudity, alcohol, tobacco, drugs, brand logos, or anything a parent would find inappropriate for a 4 year old."""
KIDS_BANNED = re.compile(
r"\b(kill\w*|blood\w*|bleed\w*|guns?|knife|knives|swords?|weapons?|bombs?|murder\w*|dead|die|dies|dying|death|"
r"scary|spooky|terrif\w*|horror|creepy|nightmare\w*|ghost\w*|haunt\w*|zombie\w*|demon\w*|devil\w*|bhoot|"
r"sexy|(?<!sun-)(?<!sun )kiss\w*|beer|wine|drunk|cigar\w*|smok\w*|drugs?|vape\w*|hate)\b",
re.I,
)
KIDS_CONTACT = re.compile(
r"(https?://|www\.|\S+@\S+\.\w+|@\w{3,}|\bcomments?\b|\bsubscrib\w*|\blike and\b|\blink in\b|"
r"\byour (name|address|phone|school|age|birthday)\b|\bdm\b|\bmessage us\b)",
re.I,
)
log_buffer = deque(maxlen=3000)
log_lock = threading.Lock()
log_seq = 0
class BufferHandler(logging.Handler):
def emit(self, record):
global log_seq
line = self.format(record)
with log_lock:
log_seq += 1
log_buffer.append((log_seq, record.levelname, line))
logger = logging.getLogger("shorts")
logger.setLevel(logging.DEBUG)
logger.propagate = False
_fmt = logging.Formatter("%(asctime)s [%(levelname)s] %(message)s")
for _h in (
RotatingFileHandler(LOG_FILE, maxBytes=5_000_000, backupCount=5, encoding="utf-8"),
logging.StreamHandler(),
BufferHandler(),
):
_h.setFormatter(_fmt)
logger.addHandler(_h)
state_lock = threading.RLock()
settings_lock = threading.RLock()
pipe_lock = threading.Lock()
pipeline = {"id": None, "obj": None}
kokoro = {"obj": None}
stop_event = threading.Event()
runner = None
pending_flow = None
pending_profile = None
status = {"running": False, "stage": "idle", "step": 0, "steps": 0, "job": "", "started": None}
class Cancelled(Exception):
pass
class VisionUnavailable(Exception):
pass
def check():
if stop_event.is_set():
raise Cancelled()
def set_stage(stage, log=True, step=None, steps=None):
status["stage"] = stage
if step is not None:
status["step"] = step
if steps is not None:
status["steps"] = steps
if log:
logger.info("Stage: %s", stage)
def reset_progress(job=""):
status["step"] = 0
status["steps"] = 0
status["job"] = job
status["started"] = datetime.now().timestamp() if job else None
def truthy(v):
return v is True or str(v).strip().lower() in ("true", "yes", "1", "pass", "safe")
def read_json(path, default):
try:
return json.loads(path.read_text(encoding="utf-8"))
except Exception:
return default
def write_json(path, data):
tmp = path.with_suffix(path.suffix + ".tmp")
tmp.write_text(json.dumps(data, indent=2), encoding="utf-8")
os.replace(tmp, path)
def coerce(defaults, key, value):
d = defaults[key]
if isinstance(d, bool):
return bool(value)
if isinstance(d, int):
return int(value)
return str(value).strip()
def resolve_storage(path_str):
p = Path(str(path_str or "videos").strip().strip('"')).expanduser()
return p if p.is_absolute() else BASE / p
def resolve_sub(s, key, fallback):
raw = str(s.get(key, "")).strip().strip('"')
if raw:
p = Path(raw).expanduser()
return p if p.is_absolute() else BASE / p
return resolve_storage(s["storage_dir"]) / fallback
def output_root(s):
return resolve_storage(s["storage_dir"]) / "output"
def published_root(s):
return resolve_storage(s["storage_dir"]) / "published"
def writable(root):
try:
root.mkdir(parents=True, exist_ok=True)
probe = root / ".write_test"
probe.write_text("ok", encoding="utf-8")
probe.unlink()
except OSError as e:
raise ValueError(f"Folder '{root}' is not writable: {e}")
return root
def validate_storage(path_str):
root = writable(resolve_storage(path_str))
(root / "output").mkdir(parents=True, exist_ok=True)
(root / "published").mkdir(parents=True, exist_ok=True)
return root
def free_gb(path):
try:
return shutil.disk_usage(path).free / 1e9
except OSError:
return None
def venv_python():
cand = BASE / ".venv" / "bin" / "python"
return str(cand) if cand.exists() else sys.executable
def pip_install(*args):
cmd = [venv_python(), "-m", "pip", "install", *args]
logger.info("Installing %s (this can take a few minutes)", " ".join(args))
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0:
logger.warning("Install failed: %s", (r.stderr or r.stdout).strip()[-500:])
return False
logger.info("Installed %s", " ".join(args))
return True
def have_module(name):
try:
__import__(name)
return True
except Exception:
return False
def ensure_module(s, module, package):
if have_module(module):
return True
if not s.get("auto_install", True):
logger.warning("%s is missing and auto install is off", package)
return False
set_stage(f"installing {package}")
return pip_install(package) and have_module(module)
def voices_dir(s):
return resolve_sub(s, "models_dir", "models") / "voices"
def find_piper_voice(s):
configured = s["piper_model"].strip()
if configured and Path(configured).expanduser().exists():
return str(Path(configured).expanduser())
vdir = voices_dir(s)
if vdir.exists():
for f in sorted(vdir.glob("*.onnx")):
return str(f)
return ""
def ensure_piper_voice(s):
voice = find_piper_voice(s)
if voice:
return voice
if not s.get("auto_install", True):
return ""
vdir = voices_dir(s)
vdir.mkdir(parents=True, exist_ok=True)
base = "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium"
set_stage("downloading narration voice")
try:
for suffix in (".onnx", ".onnx.json"):
dest = vdir / f"en_US-lessac-medium{suffix}"
if dest.exists():
continue
logger.info("Downloading narration voice%s", suffix)
with requests.get(base + suffix, stream=True, timeout=(10, 600)) as r:
r.raise_for_status()
with open(dest, "wb") as fh:
for chunk in r.iter_content(chunk_size=1 << 20):
check()
fh.write(chunk)
logger.info("Narration voice ready")
except Cancelled:
raise
except Exception as e:
logger.warning("Could not download the Piper voice: %s", e)
return ""
return find_piper_voice(s)
def piper_binary():
for cand in (BASE / ".venv" / "bin" / "piper", Path("/opt/homebrew/bin/piper")):
if cand.exists():
return str(cand)
return shutil.which("piper") or ""
KOKORO_FILES = {
"kokoro-v1.0.onnx": "https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/kokoro-v1.0.onnx",
"voices-v1.0.bin": "https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/voices-v1.0.bin",
}
def kokoro_assets(s):
vdir = voices_dir(s)
vdir.mkdir(parents=True, exist_ok=True)
paths = {}
for name, url in KOKORO_FILES.items():
dest = vdir / name
if not dest.exists() or dest.stat().st_size < 1_000_000:
set_stage(f"downloading narration model {name}")
logger.info("Downloading %s", name)
tmp = dest.with_suffix(dest.suffix + ".part")
with requests.get(url, stream=True, timeout=(10, 1200)) as r:
r.raise_for_status()
with open(tmp, "wb") as fh:
for chunk in r.iter_content(chunk_size=1 << 20):
check()
fh.write(chunk)
os.replace(tmp, dest)
logger.info("Downloaded %s", name)
paths[name] = str(dest)
return paths
def kokoro_speak(s, text, out_wav):
import soundfile as sf
from kokoro_onnx import Kokoro
with pipe_lock:
if kokoro["obj"] is None:
assets = kokoro_assets(s)
logger.info("Loading narration model")
kokoro["obj"] = Kokoro(assets["kokoro-v1.0.onnx"], assets["voices-v1.0.bin"])
logger.info("Narration model ready")
voice = s["tts_voice"].strip() or "af_heart"
if not re.fullmatch(r"[a-z]{2}_[a-z_]+", voice):
voice = "af_heart"
samples, rate = kokoro["obj"].create(text, voice=voice, speed=0.95, lang="en-us")
sf.write(str(out_wav), samples, rate)
def apply_paths(s):
models = writable(resolve_sub(s, "models_dir", "models"))
temp = writable(resolve_sub(s, "temp_dir", "tmp"))
os.environ["HF_HOME"] = str(models)
os.environ["HF_HUB_CACHE"] = str(models / "hub")
os.environ["TORCH_HOME"] = str(models / "torch")
os.environ["TMPDIR"] = str(temp)
tempfile.tempdir = str(temp)
return models, temp
def prepare_storage(s):
root = validate_storage(s["storage_dir"])
free = free_gb(root)
if free is not None and free < MIN_FREE_GB:
raise RuntimeError(f"Only {free:.1f} GB free at {root}, need at least {MIN_FREE_GB} GB")
models, temp = apply_paths(s)
logger.info("Storage: %s (%.1f GB free) | Models: %s | Temp: %s", root, free or 0, models, temp)
def finish_files(s, job_dir, job_id):
try:
if s["delete_after_publish"]:
shutil.rmtree(job_dir)
logger.info("Deleted local files for job %s after publishing", job_id)
return "deleted after publish"
dest = published_root(s) / s["profile_id"] / job_id
dest.parent.mkdir(parents=True, exist_ok=True)
shutil.move(str(job_dir), str(dest))
logger.info("Moved job %s to %s", job_id, dest)
return str(dest)
except Exception as e:
logger.error("Video was uploaded but moving or deleting local files failed: %s", e)
return str(job_dir)
def pdir(pid):
return PROFILES_DIR / pid
def valid_pid(pid):
return bool(re.fullmatch(r"[a-z0-9-]+", pid or "")) and (pdir(pid) / "profile.json").exists()
def slugify(name):
base = re.sub(r"[^a-z0-9]+", "-", name.lower()).strip("-") or "profile"
slug, i = base, 2
while pdir(slug).exists():
slug = f"{base}-{i}"
i += 1
return slug
def load_profile(pid):
p = dict(PROFILE_DEFAULTS)
p.update({k: v for k, v in read_json(pdir(pid) / "profile.json", {}).items() if k in PROFILE_DEFAULTS})
return p
def save_profile(pid, p):
pdir(pid).mkdir(parents=True, exist_ok=True)
write_json(pdir(pid) / "profile.json", p)
def profile_channel(pid):
if not (pdir(pid) / "token.json").exists():
return None
return read_json(pdir(pid) / "channel.json", {}).get("title")
def list_profiles():
return [
{"id": d.name, "name": load_profile(d.name)["name"], "channel": profile_channel(d.name)}
for d in sorted(PROFILES_DIR.iterdir())
if (d / "profile.json").exists()
]
def load_global():
with settings_lock:
g = dict(GLOBAL_DEFAULTS)
g.update({k: v for k, v in read_json(SETTINGS_FILE, {}).items() if k in GLOBAL_DEFAULTS})
if not valid_pid(g["active_profile"]):
profiles = list_profiles()
g["active_profile"] = profiles[0]["id"] if profiles else "default"
return g
def save_global(g):
with settings_lock:
write_json(SETTINGS_FILE, g)
def load_settings():
g = load_global()
s = dict(g)
s.update(load_profile(g["active_profile"]))
s["profile_id"] = g["active_profile"]
return s
def migrate():
if any((d / "profile.json").exists() for d in PROFILES_DIR.iterdir()):
return
old = read_json(SETTINGS_FILE, {})
p = dict(PROFILE_DEFAULTS)
p.update({k: v for k, v in old.items() if k in PROFILE_DEFAULTS})
save_profile("default", p)
for f in ("token.json", "channel.json"):
src = DATA / f
if src.exists():
os.replace(src, pdir("default") / f)
g = dict(GLOBAL_DEFAULTS)
g.update({k: v for k, v in old.items() if k in GLOBAL_DEFAULTS})
g["active_profile"] = "default"
save_global(g)
st = read_json(STATE_FILE, None)
if st and st.get("published") and all(isinstance(v, int) for v in st["published"].values()):
st["published"] = {"default": st["published"]}
write_json(STATE_FILE, st)
logger.info("Created Default profile from existing settings")
def load_state():
with state_lock:
st = read_json(STATE_FILE, {})
st.setdefault("processed", [])
st.setdefault("published", {})
st.setdefault("history", [])
return st
def save_state(st):
with state_lock:
write_json(STATE_FILE, st)
def update_history(job_id, **fields):
with state_lock:
st = load_state()
for h in st["history"]:
if h.get("job_id") == job_id:
h.update(fields)
save_state(st)
def published_today(pid):
return load_state()["published"].get(pid, {}).get(date.today().isoformat(), 0)
def load_creds(pid):
tf = pdir(pid) / "token.json"
if not tf.exists():
return None
creds = Credentials.from_authorized_user_file(str(tf), SCOPES)
if creds.expired and creds.refresh_token:
creds.refresh(GoogleRequest())
tf.write_text(creds.to_json(), encoding="utf-8")
return creds
def youtube_client(pid):
creds = load_creds(pid)
if not creds:
raise RuntimeError(f"No YouTube channel linked to profile '{load_profile(pid)['name']}'")
return build("youtube", "v3", credentials=creds, cache_discovery=False)
def map_videos(items):
return [
{
"id": i["id"],
"title": i["snippet"]["title"],
"channel": i["snippet"].get("channelTitle", ""),
"description": i["snippet"].get("description", "")[:1500],
"tags": i["snippet"].get("tags", [])[:20],
"language": (i["snippet"].get("defaultAudioLanguage") or i["snippet"].get("defaultLanguage") or "").lower(),
"views": int(i.get("statistics", {}).get("viewCount", 0)),
}
for i in items
]
def mostly_latin(text):
stripped = re.sub(r"#\S+|https?://\S+|@\S+", " ", str(text))
letters = [c for c in stripped if c.isalpha()]
if not letters:
return False
latin = sum(1 for c in letters if ord(c) < 0x250)
return latin / len(letters) >= 0.95
def filter_language(videos, lang, strict=False):
if not lang:
return videos
kept = []
for v in videos:
if v["language"] and not v["language"].startswith(lang):
continue
if strict and not v["language"]:
continue
if lang in LATIN_LANGS and not mostly_latin(v["title"]):
continue
kept.append(v)
if len(kept) < len(videos):
logger.info("Skipped %d source videos not in language '%s'", len(videos) - len(kept), lang)
return kept
def fetch_trending(s):
yt = youtube_client(s["profile_id"])
count = max(1, min(int(s["trending_count"]), 50))
region = s["region_code"] or "US"
lang = s["source_language"].strip().lower()
query = s["trending_query"].strip()
if query:
after = (datetime.now(timezone.utc) - timedelta(days=max(1, int(s["trending_days"])))).strftime("%Y-%m-%dT%H:%M:%SZ")
params = {
"part": "id",
"q": query,
"type": "video",
"order": "viewCount",
"publishedAfter": after,
"safeSearch": "strict" if s["made_for_kids"] else "moderate",
"regionCode": region,
"maxResults": count,
}
if lang:
params["relevanceLanguage"] = lang
if s["category_id"]:
params["videoCategoryId"] = s["category_id"]
ids = [i["id"]["videoId"] for i in yt.search().list(**params).execute().get("items", []) if i.get("id", {}).get("videoId")]
items = yt.videos().list(part="snippet,statistics", id=",".join(ids)).execute().get("items", []) if ids else []
logger.info("Search '%s' returned %d videos from the last %s days (region %s, language %s)", query, len(items), s["trending_days"], region, lang or "any")
else:
params = {
"part": "snippet,statistics",
"chart": "mostPopular",
"regionCode": region,
"maxResults": count,
}
if s["category_id"]:
params["videoCategoryId"] = s["category_id"]
items = yt.videos().list(**params).execute().get("items", [])
logger.info("Fetched %d trending chart videos (region %s, category %s)", len(items), region, s["category_id"] or "all")
kept = filter_language(map_videos(items), lang, strict=bool(s["made_for_kids"]))
if not kept and s["made_for_kids"]:
logger.info("No sources declared language '%s', falling back to title matching", lang)
kept = filter_language(map_videos(items), lang)
return kept
def ollama_tags(s):
try:
r = requests.get(f"{s['ollama_url'].rstrip('/')}/api/tags", timeout=10)
r.raise_for_status()
return sorted(m["name"] for m in r.json().get("models", []) if m.get("name"))
except Exception as e:
logger.debug("Could not list Ollama models: %s", e)
return []
def ollama_has(s, model):
installed = ollama_tags(s)
return model in installed or f"{model}:latest" in installed
def ollama_pull(s, model):
model = (model or "").strip()
if not model:
raise RuntimeError("No model name given")
if ollama_has(s, model):
return False
logger.info("Model '%s' is not installed, pulling it now (this can take a while)", model)
url = f"{s['ollama_url'].rstrip('/')}/api/pull"
last = ""
with requests.post(url, json={"model": model, "stream": True}, stream=True, timeout=(10, 3600)) as r:
r.raise_for_status()
for line in r.iter_lines():
check()
if not line:
continue
try:
msg = json.loads(line)
except ValueError:
continue
if msg.get("error"):
raise RuntimeError(f"Could not pull '{model}': {msg['error']}")
st = msg.get("status", "")
total, done = msg.get("total"), msg.get("completed")
if total and done:
pct = int(done * 100 / total)
line_txt = f"{st} {pct}%"
if line_txt != last and pct % 10 == 0:
logger.info("Pulling %s: %s", model, line_txt)
last = line_txt
elif st and st != last:
logger.info("Pulling %s: %s", model, st)
last = st
logger.info("Model '%s' is ready", model)
return True
def ensure_local_deps(s):
if not ensure_module(s, "torch", "torch") or not ensure_module(s, "diffusers", "diffusers accelerate safetensors transformers".split()[0]):
raise RuntimeError("Could not install the image packages. Run ./install-local.sh")
for module, package in (("accelerate", "accelerate"), ("safetensors", "safetensors"), ("transformers", "transformers")):
ensure_module(s, module, package)
want = s["tts_engine"] if s["tts_engine"] in TTS_ENGINES else "auto"
if want in ("auto", "kokoro") and not have_module("kokoro_onnx"):
if ensure_module(s, "kokoro_onnx", "kokoro-onnx"):
ensure_module(s, "soundfile", "soundfile")
else:
logger.warning("Kokoro could not be installed, trying Piper instead")
if not piper_binary():
ensure_module(s, "piper", "piper-tts")
ensure_piper_voice(s)
if want == "piper":
if not piper_binary():
ensure_module(s, "piper", "piper-tts")
ensure_piper_voice(s)
logger.info("Narration engine: %s", pick_tts(s))
def ensure_models(s, engine):
needed = [s["ollama_model"].strip()]
if s["made_for_kids"] and s["vision_model"].strip():
needed.append(s["vision_model"].strip())
for m in needed:
if m and not ollama_has(s, m):
set_stage(f"downloading model {m}")
ollama_pull(s, m)
def ollama_call(s, prompt, temperature=0.9, model=None, images=None, num_predict=2048, as_json=True):
url = f"{s['ollama_url'].rstrip('/')}/api/generate"
model = model or s["ollama_model"]
timeout = max(30, int(s["ollama_timeout"]))
payload = {
"model": model,
"prompt": prompt,
"stream": False,
"keep_alive": "5m",
"options": {"temperature": temperature, "num_predict": num_predict, "num_ctx": 8192},
}
if as_json:
payload["format"] = "json"
if images:
payload["images"] = images
started = datetime.now()
logger.debug("Waiting on Ollama %s (timeout %ds)", model, timeout)
try:
r = requests.post(url, json=payload, timeout=(10, timeout))
except requests.exceptions.ReadTimeout:
raise RuntimeError(f"Ollama model '{model}' did not respond within {timeout}s. Check 'ollama ps' and free RAM, or use a smaller model")
except requests.exceptions.ConnectionError:
raise RuntimeError(f"Cannot reach Ollama at {s['ollama_url']}. Is it running?")
if r.status_code == 404:
logger.warning("Ollama model '%s' is missing, pulling it now", model)
ollama_pull(s, model)
r = requests.post(url, json=payload, timeout=(10, timeout))
r.raise_for_status()
text = r.json().get("response", "")
logger.debug("Ollama %s responded in %.1fs: %s", model, (datetime.now() - started).total_seconds(), text[:2000])
return json.loads(text) if as_json else text
def ollama_json(s, prompt, temperature=0.9, model=None, images=None, num_predict=2048):
return ollama_call(s, prompt, temperature=temperature, model=model, images=images, num_predict=num_predict, as_json=True)
def write_script(s, src, feedback=None):
clip = int(s["clip_seconds"])
n = max(1, int(s["target_seconds"]) // clip)
kids = bool(s["made_for_kids"])
words = max(8, int(clip * (2.0 if kids else 2.3)))
niche = s["channel_niche"].strip()
niche_block = (
f"This channel's niche: {niche}\nThe Short must fit this niche even if the trending source does not. Borrow only the general theme or what makes it appealing.\n"
if niche else ""
)
kids_block = KIDS_RULES if kids else ""
feedback_block = (
"A previous draft was rejected for these problems. Do not repeat them:\n" + "\n".join(f"- {f}" for f in feedback) + "\n"
if feedback else ""
)
prompt = f"""You are a short-form video writer. This video is currently popular on YouTube:
Title: {src['title']}
Channel: {src['channel']}
Description: {src['description']}
Tags: {', '.join(src['tags'])}
{niche_block}{kids_block}{feedback_block}
Identify the underlying topic and why it appeals to viewers. Then write a completely ORIGINAL {n * clip}-second YouTube Short in English.
Do not reuse the source's title, script, characters, jokes, branding, or channel identity.
Do not depict real people, celebrities, brands, logos, or copyrighted characters. Invent new characters.
First invent the cast. "characters" is one sentence naming each character with fixed, concrete visual details (species, color, size, clothing, one distinctive feature) that never change.
Every character is a cartoon: a cartoon animal, a cartoon creature, or a simple cartoon person. Never describe anyone as realistic, lifelike, photorealistic, or human-looking.
The Short has exactly {n} scenes of {clip} seconds each. Each scene has:
"visual": a detailed, self-contained shot description of one still image (setting, action, mood, lighting). Name the characters but do not re-describe their appearance, that comes from "characters". Never mention on-screen text or words.
"narration": one spoken line of at most {words} words.
Scene 1 must hook the viewer in the first 2 seconds. The final scene must deliver a payoff.
Return JSON only in this shape:
{{"title": "under 70 characters", "characters": "one sentence describing every character's fixed appearance", "style": "one sentence visual style applied to every scene", "description": "2 to 3 sentence YouTube description", "scenes": [{{"visual": "...", "narration": "..."}}]}}"""
last = None
for attempt in range(1, 4):
check()
try:
data = ollama_json(s, prompt)
scenes = [
sc for sc in data.get("scenes", [])
if isinstance(sc, dict) and str(sc.get("visual", "")).strip()
]
if len(scenes) < n:
raise ValueError(f"expected {n} scenes, got {len(scenes)}")
if not str(data.get("title", "")).strip():
raise ValueError("missing title")
data["scenes"] = [{"visual": str(sc.get("visual", "")), "narration": str(sc.get("narration", ""))} for sc in scenes[:n]]
data["title"] = str(data["title"])
data["style"] = str(data.get("style", ""))
data["characters"] = str(data.get("characters", ""))
data["description"] = str(data.get("description", ""))
logger.info("Script ready: '%s' (%d scenes)", data["title"], n)
return data
except Cancelled:
raise
except Exception as e:
last = e
logger.warning("Script attempt %d failed: %s", attempt, e)
raise RuntimeError(f"Could not generate a valid script: {last}")
def clean_tags(tags):
out, seen, total = [], set(), 0
for t in tags:
t = re.sub(r"[<>#\"]", "", str(t)).strip()
if not t or len(t) > 100 or t.lower() in seen:
continue
cost = len(t) + (2 if " " in t else 0) + (1 if out else 0)
if total + cost > 480:
break
out.append(t)
seen.add(t.lower())
total += cost
return out
def make_tags(s, script):
narration = " ".join(sc["narration"] for sc in script["scenes"])
niche = s["channel_niche"].strip()
niche_line = f"Channel niche: {niche}\n" if niche else ""
kids_line = "The audience is young children. Tags must be child-appropriate and describe kids content.\n" if s["made_for_kids"] else ""
prompt = f"""Generate YouTube tags for this Short.
Include specific tags for its exact subject plus broader category, genre, and audience tags that help discovery.
Every tag must accurately describe the video. No unrelated trending terms, no names of other creators, people, brands, or copyrighted characters.
{niche_line}{kids_line}
Title: {script['title']}
Description: {script['description']}
Narration: {narration}
Return JSON only: {{"tags": ["15 to 25 tags, most specific first"]}}"""
for attempt in range(1, 3):
check()
try:
data = ollama_json(s, prompt, num_predict=512)
raw = data.get("tags", [])
if isinstance(raw, str):
raw = raw.split(",")
tags = clean_tags(raw)
if tags:
return tags
except Cancelled:
raise
except Exception as e:
logger.warning("Tag attempt %d failed: %s", attempt, e)
logger.warning("No tags generated, uploading without tags")
return []
def normalize_text(t):
return re.sub(r"\s+", " ", re.sub(r"[^a-z0-9 ]", " ", str(t).lower())).strip()
def script_text(script):
parts = [script["title"], script["description"], script["style"], script.get("characters", "")]
for sc in script["scenes"]:
parts += [sc["visual"], sc["narration"]]
return " ".join(parts)
def kids_rule_issues(script):
issues = []
texts = [script["title"], script["description"], script["style"], script.get("characters", "")]
texts += [f"{sc['visual']} {sc['narration']}" for sc in script["scenes"]]
for t in texts:
for m in KIDS_BANNED.finditer(t):
issues.append(f"banned word '{m.group(0)}'")
for m in KIDS_CONTACT.finditer(t):
issues.append(f"contact info or call to action '{m.group(0)}'")
title = script["title"]
if re.search(r"\b[A-Z]{4,}\b", title):
issues.append("ALL CAPS word in title")
if title.count("!") > 1 or title.count("?") > 1:
issues.append("clickbait punctuation in title")
return sorted(set(issues))
def kids_llm_review(s, script):
scenes = "\n".join(f"{i}. Visual: {sc['visual']}\n Narration: {sc['narration']}" for i, sc in enumerate(script["scenes"], 1))
rules = "\n".join(f"{k}: {v}" for k, v in REVIEW_RULES.items())
prompt = f"""You are a careful reviewer for a YouTube channel of cartoon stories for young children.
Check the script below against each rule. A rule is only broken if specific words in the script clearly break it.
Do not judge capitalization, punctuation, target age, art style, or anything not listed in the rules. Those are checked elsewhere.
Rules (id: what is not allowed):
{rules}
Title: {script['title']}
Description: {script['description']}
Visual style: {script['style']}
Scenes:
{scenes}
For every broken rule, copy the exact words from the script that break it.
Return JSON only: {{"violations": [{{"rule": "rule id", "quote": "exact words copied from the script", "reason": "short explanation"}}]}}
Return {{"violations": []}} if no rule is broken."""
try:
data = ollama_json(s, prompt, temperature=0.1, num_predict=1024)
except Cancelled:
raise
except Exception as e:
return [f"compliance reviewer error: {e}"], 0
raw = data.get("violations", [])
if not isinstance(raw, list):
return ["compliance reviewer returned an invalid response"], 0
haystack = normalize_text(script_text(script))
verified, ignored = [], 0
for v in raw:
if not isinstance(v, dict):
ignored += 1
continue
rule = str(v.get("rule", "")).strip().lower()
quote = normalize_text(v.get("quote", ""))
if rule not in REVIEW_RULES or len(quote) < 3 or quote not in haystack:
ignored += 1
logger.debug("Ignored unverified reviewer finding: %s", v)
continue
verified.append(f"{rule}: '{v.get('quote')}' ({v.get('reason', '')})")
return verified, ignored
def review_mode(s):
mode = s["kids_llm_review"] if s["kids_llm_review"] in REVIEW_MODES else "advisory"
if mode == "advisory" and not s["kids_manual_review"]:
return "block"
return mode
def produce_script(s, src):
kids = bool(s["made_for_kids"])
feedback = []
attempts = 4 if kids else 1
mode = review_mode(s)
if kids and mode != s["kids_llm_review"]:
logger.info("LLM review switched to block mode because manual review is off")
for attempt in range(1, attempts + 1):
check()
set_stage("writing script")
script = write_script(s, src, feedback)
set_stage("generating tags")
tags = make_tags(s, script)
if not kids:
logger.info("Tags: %s", ", ".join(tags))
return script, tags, {"kids_checks": False}
dropped = [t for t in tags if KIDS_BANNED.search(t) or KIDS_CONTACT.search(t)]
tags = [t for t in tags if t not in dropped]
if dropped:
logger.info("Dropped tags that failed kids filters: %s", ", ".join(dropped))
logger.info("Tags: %s", ", ".join(tags))
set_stage("kids compliance review")
issues = kids_rule_issues(script)
notes, ignored = ([], 0) if mode == "off" else kids_llm_review(s, script)
if ignored:
logger.info("Ignored %d reviewer findings that did not quote the script or match a rule", ignored)
if mode == "block":
issues += notes
if not issues:
if notes:
logger.warning("Reviewer notes (advisory, check during manual review): %s", "; ".join(notes))
logger.info("Kids compliance review passed on attempt %d (LLM review: %s)", attempt, mode)
return script, tags, {
"kids_checks": True,
"script_review": "passed",
"attempts": attempt,
"dropped_tags": dropped,
"reviewer_mode": mode,
"reviewer_notes": notes,
"reviewer_ignored_findings": ignored,
}
logger.warning("Kids compliance review failed (attempt %d/%d): %s", attempt, attempts, "; ".join(issues))
feedback = issues
raise RuntimeError(f"Script failed kids compliance review after {attempts} attempts")
def unload_pipeline():
with pipe_lock:
if pipeline["obj"] is None:
return
pipeline["obj"] = None
pipeline["id"] = None
try:
import gc
import torch
gc.collect()
if torch.backends.mps.is_available():
torch.mps.empty_cache()
elif torch.cuda.is_available():
torch.cuda.empty_cache()
except Exception:
pass
logger.debug("Released image model from memory")
def kids_vision_check(s, clip_path):
model = s["vision_model"].strip()
if not model:
raise RuntimeError("Made for kids requires a vision model in Settings to check generated clips")
frame = clip_path.with_name(f"{clip_path.stem}_frame.jpg")
r = subprocess.run(
["ffmpeg", "-y", "-ss", str(int(s["clip_seconds"]) / 2), "-i", str(clip_path), "-frames:v", "1", "-vf", "scale=512:-2", str(frame)],
capture_output=True, text=True,
)
if r.returncode != 0 or not frame.exists():
raise RuntimeError("Could not extract a frame for the vision check")
images = [base64.b64encode(frame.read_bytes()).decode()]
frame.unlink(missing_ok=True)
if not model.split(":")[0] in SMALL_VISION_MODELS:
unload_pipeline()
safe, reason = True, ""
try:
try:
data = ollama_call(s, KIDS_VISION_PROMPT, temperature=0.1, model=model, images=images, num_predict=256)
safe = truthy(data.get("safe"))
reason = str(data.get("reason", ""))
except (requests.exceptions.HTTPError, ValueError):
text = ollama_call(s, VISION_TEXT_PROMPT, temperature=0.1, model=model, images=images, num_predict=128, as_json=False)
low = text.lower()
safe = "unsafe" not in low
reason = text.strip()[:200]
except Cancelled:
raise
except Exception as e:
raise VisionUnavailable(f"vision model '{model}' could not run: {e}") from e
if not safe:
raise ValueError(f"Vision check rejected {clip_path.name}: {reason or 'no reason given'}")
logger.info("Vision check passed for %s", clip_path.name)
def load_pipeline(s):
with pipe_lock:
model = s["sd_model"].strip()
if pipeline["id"] == model and pipeline["obj"] is not None:
return pipeline["obj"]
try:
import torch
from diffusers import AutoPipelineForText2Image
except ImportError:
raise RuntimeError("Image packages are missing. Turn on auto install in Settings or run ./install-local.sh")
if torch.backends.mps.is_available():
device, dtype = "mps", torch.float16
elif torch.cuda.is_available():
device, dtype = "cuda", torch.float16
else:
device, dtype = "cpu", torch.float32
logger.info("Loading image model %s on %s (first run downloads several GB)", model, device)
pipe = AutoPipelineForText2Image.from_pretrained(model, torch_dtype=dtype, variant="fp16" if dtype == torch.float16 else None)
pipe = pipe.to(device)
pipe.set_progress_bar_config(disable=True)
pipeline["id"] = model
pipeline["obj"] = pipe
logger.info("Image model ready")
return pipe
def generate_image(s, prompt, path):
pipe = load_pipeline(s)
kwargs = {
"prompt": prompt[:900],
"num_inference_steps": max(1, int(s["sd_steps"])),
"guidance_scale": float(s["sd_guidance"]),
"width": max(512, int(s["sd_width"]) // 64 * 64),
"height": max(512, int(s["sd_height"]) // 64 * 64),
}
if kwargs["guidance_scale"] < 1.1:
kwargs["guidance_scale"] = 1.5
kwargs["negative_prompt"] = IMAGE_NEGATIVE
started = datetime.now()
image = pipe(**kwargs).images[0]
image.save(str(path))
logger.info("Generated %s in %.1fs", path.name, (datetime.now() - started).total_seconds())
def pick_tts(s):
engine = s["tts_engine"] if s["tts_engine"] in TTS_ENGINES else "auto"
if engine != "auto":
return engine
if have_module("kokoro_onnx"):
return "kokoro"
if find_piper_voice(s) and piper_binary():
return "piper"
return "say" if sys.platform == "darwin" else "espeak"
def system_speak(s, text, out_wav):
if sys.platform == "darwin":
aiff = out_wav.with_suffix(".aiff")
cmd = ["say", "-r", str(int(s["tts_rate"])), "-o", str(aiff), text]
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0:
raise RuntimeError(f"say failed: {r.stderr.strip()[:200]}")
conv = subprocess.run(["ffmpeg", "-y", "-i", str(aiff), "-ar", "44100", "-ac", "2", str(out_wav)], capture_output=True, text=True)
aiff.unlink(missing_ok=True)
if conv.returncode != 0:
raise RuntimeError("Could not convert narration audio")
else:
r = subprocess.run(["espeak-ng", "-v", "en-us", "-s", str(int(s["tts_rate"])), "-w", str(out_wav), text], capture_output=True, text=True)
if r.returncode != 0:
raise RuntimeError(f"espeak-ng failed: {r.stderr.strip()[:200]}")
def tts_speak(s, text, out_wav):
text = text.strip() or "..."
engine = pick_tts(s)
try:
if engine == "kokoro":
kokoro_speak(s, text, out_wav)
elif engine == "piper":
voice = ensure_piper_voice(s)
binary = piper_binary()
if not voice or not binary:
raise RuntimeError("Piper is not available")
r = subprocess.run([binary, "--model", voice, "--output_file", str(out_wav)], input=text, capture_output=True, text=True)
if r.returncode != 0:
raise RuntimeError(f"piper failed: {r.stderr.strip()[:200]}")
else:
system_speak(s, text, out_wav)
except Cancelled:
raise
except Exception as e:
if engine in ("kokoro", "piper"):
logger.warning("%s narration failed (%s), using the system voice for this scene", engine, e)
system_speak(s, text, out_wav)
else:
raise
if not out_wav.exists() or out_wav.stat().st_size < 1000:
raise RuntimeError("Narration audio was empty")
def media_duration(path):
r = subprocess.run(
["ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", str(path)],
capture_output=True, text=True,
)
try:
return float(r.stdout.strip())
except ValueError:
return 0.0
def render_scene(s, image, audio, out, aspect):
w, h = (1080, 1920) if aspect == "9:16" else (1920, 1080)
dur = max(float(s["clip_seconds"]), media_duration(audio) + 0.6)
frames = int(dur * 30)
vf = (
f"[0:v]scale={w * 2}:{h * 2}:force_original_aspect_ratio=increase,crop={w * 2}:{h * 2},"
f"zoompan=z='min(zoom+0.0004,1.15)':x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)':d={frames}:s={w}x{h}:fps=30,"
f"format=yuv420p[v];[1:a]apad,atrim=0:{dur:.2f},asetpts=N/SR/TB[a]"
)
cmd = [
"ffmpeg", "-y", "-loop", "1", "-i", str(image), "-i", str(audio),
"-filter_complex", vf, "-map", "[v]", "-map", "[a]",
"-c:v", "libx264", "-preset", "medium", "-crf", "20",
"-c:a", "aac", "-b:a", "160k", "-t", f"{dur:.2f}", str(out),
]
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0:
logger.error("ffmpeg output:\n%s", r.stderr[-3000:])
raise RuntimeError("Could not render scene video")
logger.info("Rendered %s (%.1fs)", out.name, dur)
def make_clip_local(s, scene, script, path, kids):
image = path.with_suffix(".png")
audio = path.with_suffix(".wav")
style = script["style"].strip()
cast = script.get("characters", "").strip()
cast_part = f" Characters: {cast}" if cast else ""
prompt = f"{scene['visual']}{cast_part} {style}{KIDS_IMAGE_SUFFIX if kids else ''}{CARTOON_SUFFIX}"
logger.debug("Image prompt: %s", prompt)
generate_image(s, prompt, image)
tts_speak(s, scene["narration"], audio)
render_scene(s, image, audio, path, s["aspect_ratio"])
if not s["keep_clips"]:
audio.unlink(missing_ok=True)
return image
def make_clip_veo(s, scene, script, path, kids):
from google import genai
from google.genai import types
client = genai.Client(api_key=s["gemini_api_key"])
orientation = "Vertical" if s["aspect_ratio"] == "9:16" else "Widescreen"
prompt = (
f"{scene['visual']} Style: {script['style']}. {orientation} short-form video. "
f"A narrator's voiceover says: \"{scene['narration']}\" "
f"No on-screen text, captions, or subtitles.{KIDS_VEO_SUFFIX if kids else ''}"
)
logger.debug("Veo prompt: %s", prompt)
op = client.models.generate_videos(
model=s["veo_model"],
prompt=prompt,
config=types.GenerateVideosConfig(aspect_ratio=s["aspect_ratio"]),
)
waited = 0
while not op.done:
check()
if waited > 900:
raise TimeoutError("Veo generation timed out after 15 minutes")
stop_event.wait(10)
waited += 10
op = client.operations.get(op)
err = getattr(op, "error", None)
if err:
raise RuntimeError(f"Veo error: {err}")
vids = op.response.generated_videos if op.response else None
if not vids:
raise RuntimeError("Veo returned no video (prompt may have been blocked by safety filters)")
v = vids[0]
client.files.download(file=v.video)
v.video.save(str(path))
logger.info("Saved clip %s", path.name)
return None
def concat_clips(clips, out, aspect):
lst = out.with_suffix(".txt")
lst.write_text("".join(f"file '{c.resolve().as_posix()}'\n" for c in clips), encoding="utf-8")
w, h = (1080, 1920) if aspect == "9:16" else (1920, 1080)
cmd = [
"ffmpeg", "-y", "-f", "concat", "-safe", "0", "-i", str(lst),
"-vf", f"scale={w}:{h}:force_original_aspect_ratio=decrease,pad={w}:{h}:(ow-iw)/2:(oh-ih)/2,fps=30",
"-c:v", "libx264", "-preset", "medium", "-crf", "20",
"-c:a", "aac", "-b:a", "160k", "-movflags", "+faststart", str(out),
]
logger.debug("ffmpeg: %s", " ".join(cmd))
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0:
logger.error("ffmpeg output:\n%s", r.stderr[-3000:])
raise RuntimeError(f"ffmpeg failed with code {r.returncode}")
lst.unlink(missing_ok=True)
logger.info("Final video: %s", out)
def upload_video(s, path, script, tags):
yt = youtube_client(s["profile_id"])
kids = bool(s["made_for_kids"])
title = re.sub(r"[<>]", "", script["title"]).strip()[:90] or "Untitled"
if "#shorts" not in title.lower():
title = f"{title} #Shorts"
hashtags = []
for t in tags[:3]:
h = re.sub(r"[^A-Za-z0-9]", "", t)
if h:
hashtags.append("#" + h)
description = re.sub(r"[<>]", "", f"{script['description']}\n\n#Shorts {' '.join(hashtags)}").strip()[:4500]
privacy = s["privacy_status"]
held = kids and s["kids_manual_review"]
if held:
privacy = "private"
logger.info("Holding made-for-kids video as private for manual review")
body = {
"snippet": {
"title": title,
"description": description,
"tags": tags,
"categoryId": s["upload_category_id"] or "24",
},
"status": {
"privacyStatus": privacy,
"selfDeclaredMadeForKids": kids,
"containsSyntheticMedia": True,
},
}
logger.info("Uploading to %s as category %s, made for kids: %s", profile_channel(s["profile_id"]), body["snippet"]["categoryId"], kids)
media = MediaFileUpload(str(path), mimetype="video/mp4", resumable=True, chunksize=8 * 1024 * 1024)
req = yt.videos().insert(part="snippet,status", body=body, media_body=media)
resp = None
while resp is None:
check()
prog, resp = req.next_chunk()
if prog:
logger.info("Upload %d%%", int(prog.progress() * 100))
logger.info("Uploaded: https://youtube.com/shorts/%s (%s)", resp["id"], privacy)
return resp["id"], held
def run_job(s):
engine = s["video_engine"] if s["video_engine"] in ENGINES else "local"
if engine == "veo" and not s["gemini_api_key"]:
raise RuntimeError("Gemini API key is not set in Settings")
if not shutil.which("ffmpeg"):
raise RuntimeError("ffmpeg was not found on PATH")
kids = bool(s["made_for_kids"])
if kids and not s["vision_model"].strip():
raise RuntimeError("Made for kids requires a vision model in Settings (e.g. llama3.2-vision)")
prepare_storage(s)
ensure_models(s, engine)
if engine == "local":
ensure_local_deps(s)
if engine == "local":
logger.info("Image model: %s at %sx%s", s["sd_model"], s["sd_width"], s["sd_height"])
logger.info("Profile: %s | Engine: %s | Niche: %s | Made for kids: %s | Upload category: %s", s["name"], engine, s["channel_niche"] or "none", kids, s["upload_category_id"])
if kids:
logger.info(
"Kids compliance active: script rules, word filters, LLM review (%s), frame vision checks, made-for-kids flag%s",
review_mode(s),
", manual review hold" if s["kids_manual_review"] else "",
)
set_stage("fetching trending videos")
trending = fetch_trending(s)
state = load_state()
candidates = [v for v in sorted(trending, key=lambda v: -v["views"]) if v["id"] not in state["processed"]]
if kids:
safe = [v for v in candidates if not KIDS_BANNED.search(v["title"])]
if len(safe) < len(candidates):
logger.info("Skipped %d source videos with themes unsuitable for kids", len(candidates) - len(safe))
candidates = safe
if not candidates:
logger.warning("No usable trending videos, checking again in 30 minutes")
set_stage("waiting for new trending videos")
stop_event.wait(1800)
return
src = candidates[0]
job_id = datetime.now().strftime("%Y%m%d-%H%M%S")
job_dir = output_root(s) / s["profile_id"] / job_id
job_dir.mkdir(parents=True, exist_ok=True)
with state_lock:
st = load_state()
st["processed"] = (st["processed"] + [src["id"]])[-2000:]
st["history"] = ([{
"job_id": job_id,
"profile": s["name"],
"started": datetime.now().strftime("%Y-%m-%d %H:%M"),
"source_id": src["id"],
"source_title": src["title"],
"status": "running",
"location": str(job_dir),
}] + st["history"])[:200]
save_state(st)
scenes_n = max(1, int(s["target_seconds"]) // max(1, int(s["clip_seconds"])))
reset_progress(job_id)
set_stage(status["stage"], log=False, step=1, steps=3 + scenes_n * 2 + 2)
logger.info("Job %s: source %s '%s' (%s views)", job_id, src["id"], src["title"], f"{src['views']:,}")
logger.info("Working folder: %s", job_dir)
compliance = {}
finalized = False
try:
script, tags, compliance = produce_script(s, src)
compliance["engine"] = engine
update_history(job_id, title=script["title"], notes="; ".join(compliance.get("reviewer_notes", []))[:400])
write_json(job_dir / "script.json", {"profile": s["name"], "source": src, "script": script, "tags": tags})
clips = []
extras = []
vision_log = []
n = len(script["scenes"])
for i, scene in enumerate(script["scenes"], 1):
check()
set_stage(f"generating scene {i}/{n}", step=3 + (i - 1) * 2)
path = job_dir / f"clip_{i:02d}.mp4"
for attempt in range(1, 4):
try:
if engine == "local":
extra = make_clip_local(s, scene, script, path, kids)
else:
extra = make_clip_veo(s, scene, script, path, kids)
except Cancelled:
raise
except Exception as e:
logger.warning("Scene %d attempt %d failed: %s", i, attempt, e)
path.unlink(missing_ok=True)
if attempt == 3:
raise
stop_event.wait(10)
continue
if kids:
set_stage(f"vision check scene {i}/{n}", step=4 + (i - 1) * 2)
try:
kids_vision_check(s, path)
vision_log.append({"clip": i, "attempt": attempt, "result": "passed"})
except Cancelled:
raise
except VisionUnavailable as e:
if s["kids_manual_review"]:
logger.warning("Scene %d: %s. Continuing, check this video manually before publishing", i, e)
vision_log.append({"clip": i, "attempt": attempt, "result": f"skipped: {e}"[:300]})
else:
raise RuntimeError(f"{e}. Turn on manual review or pick a working vision model")
except Exception as e:
logger.warning("Scene %d attempt %d rejected: %s", i, attempt, e)
vision_log.append({"clip": i, "attempt": attempt, "result": str(e)[:300]})
path.unlink(missing_ok=True)
if attempt == 3:
raise
stop_event.wait(5)
continue
if extra:
extras.append(extra)
break
clips.append(path)
if kids:
compliance["vision_checks"] = vision_log
set_stage("stitching video", step=3 + n * 2)
final = job_dir / "final.mp4"
concat_clips(clips, final, s["aspect_ratio"])
set_stage("uploading to YouTube", step=4 + n * 2)
vid, held = upload_video(s, final, script, tags)
compliance["made_for_kids_flag"] = kids
compliance["synthetic_media_flag"] = True
compliance["held_for_review"] = held
compliance["youtube_id"] = vid
with state_lock:
st = load_state()
key = date.today().isoformat()
pub = st["published"].setdefault(s["profile_id"], {})
pub[key] = pub.get(key, 0) + 1
st["published"][s["profile_id"]] = dict(sorted(pub.items())[-60:])
save_state(st)
write_json(job_dir / "compliance.json", compliance)
finalized = True
if not s["keep_clips"]:
for c in clips + extras:
c.unlink(missing_ok=True)
set_stage("moving files")
location = finish_files(s, job_dir, job_id)
update_history(job_id, status="review" if held else "published", youtube_id=vid, location=location)
set_stage("job complete", step=status["steps"])
except Cancelled:
update_history(job_id, status="cancelled")
logger.warning("Job %s cancelled", job_id)
raise
except Exception as e:
update_history(job_id, status="failed", error=str(e)[:300])
raise
finally:
if compliance and not finalized and job_dir.exists():
write_json(job_dir / "compliance.json", compliance)
def run_loop():
status["running"] = True
logger.info("Autopilot started")
try:
while not stop_event.is_set():
s = load_settings()
limit = int(s["videos_per_day"])
if limit > 0 and published_today(s["profile_id"]) >= limit:
if status["stage"] != "daily limit reached":
logger.info("Profile '%s' reached its daily limit of %d, waiting for tomorrow", s["name"], limit)
set_stage("daily limit reached", log=False)
stop_event.wait(60)
continue
try:
run_job(s)
except Cancelled:
break
except Exception as e:
logger.exception("Job failed: %s", e)
mins = max(1, int(s["error_cooldown_minutes"]))
set_stage(f"cooling down {mins} min after error")
stop_event.wait(mins * 60)
continue
gap = int(s["minutes_between_videos"])
if gap > 0 and not stop_event.is_set():
set_stage(f"waiting {gap} min before next video")
stop_event.wait(gap * 60)
finally:
status["running"] = False
reset_progress()
set_stage("idle")
logger.info("Autopilot stopped")
migrate()
app = FastAPI()
@app.get("/")
def index():
return FileResponse(BASE / "static" / "index.html")
@app.get("/api/status")
def api_status():
s = load_settings()
st = load_state()
root = resolve_storage(s["storage_dir"])
free = free_gb(root) if root.exists() else None
return {
"running": bool(runner and runner.is_alive()),
"stage": status["stage"],
"step": status["step"],
"steps": status["steps"],
"job": status["job"],
"elapsed": int(datetime.now().timestamp() - status["started"]) if status["started"] else 0,
"profile": s["name"],
"engine": s["video_engine"],
"made_for_kids": s["made_for_kids"],
"published_today": published_today(s["profile_id"]),
"videos_per_day": s["videos_per_day"],
"channel": profile_channel(s["profile_id"]),
"storage": str(root),
"storage_free_gb": round(free, 1) if free is not None else None,
"delete_after_publish": s["delete_after_publish"],
"history": st["history"][:25],
}
@app.get("/api/settings")
def get_settings():
g = load_global()
if g["gemini_api_key"]:
g["gemini_api_key"] = MASK
return {
"global": g,
"profile": load_profile(g["active_profile"]),
"active": g["active_profile"],
"profiles": list_profiles(),
"storage_resolved": str(resolve_storage(g["storage_dir"])),
"models_resolved": str(resolve_sub(g, "models_dir", "models")),
"temp_resolved": str(resolve_sub(g, "temp_dir", "tmp")),
"installed_models": ollama_tags(g),
"suggested_text_models": SUGGESTED_TEXT_MODELS,
"suggested_vision_models": SUGGESTED_VISION_MODELS,
}
@app.post("/api/settings")
async def post_settings(request: Request):
body = await request.json()
with settings_lock:
g = load_global()
old_storage = g["storage_dir"]
pid = g["active_profile"]
p = load_profile(pid)
try:
for k, v in body.get("global", {}).items():
if k not in GLOBAL_DEFAULTS or k == "active_profile" or (k == "gemini_api_key" and v == MASK):
continue
g[k] = coerce(GLOBAL_DEFAULTS, k, v)
for k, v in body.get("profile", {}).items():
if k in PROFILE_DEFAULTS:
p[k] = coerce(PROFILE_DEFAULTS, k, v)
g["storage_dir"] = g["storage_dir"].strip().strip('"') or "videos"
if g["video_engine"] not in ENGINES:
g["video_engine"] = "local"
if g["tts_engine"] not in TTS_ENGINES:
g["tts_engine"] = "auto"
p["source_language"] = p["source_language"].lower()[:5]
if p["kids_llm_review"] not in REVIEW_MODES:
p["kids_llm_review"] = "advisory"
root = validate_storage(g["storage_dir"])
writable(resolve_sub(g, "models_dir", "models"))
writable(resolve_sub(g, "temp_dir", "tmp"))
except (TypeError, ValueError) as e:
return JSONResponse({"detail": str(e)}, status_code=400)
if not p["name"]:
p["name"] = pid
save_global(g)
save_profile(pid, p)
if g["storage_dir"] != old_storage:
logger.info("Storage folder set to %s (applies to new jobs)", root)
logger.info("Settings saved for profile '%s'", p["name"])
return {"ok": True}
@app.post("/api/profiles")
async def create_profile(request: Request):
body = await request.json()
name = str(body.get("name", "")).strip()
if not name:
return JSONResponse({"detail": "Profile name is required"}, status_code=400)
with settings_lock:
g = load_global()
p = load_profile(g["active_profile"])
p["name"] = name
pid = slugify(name)
save_profile(pid, p)
g["active_profile"] = pid
save_global(g)
logger.info("Created profile '%s' and made it active", name)
return {"ok": True, "id": pid}
@app.post("/api/profiles/activate")
async def activate_profile(request: Request):
pid = (await request.json()).get("id", "")
if not valid_pid(pid):
return JSONResponse({"detail": "Unknown profile"}, status_code=400)
with settings_lock:
g = load_global()
g["active_profile"] = pid
save_global(g)
suffix = " (applies from the next video)" if runner and runner.is_alive() else ""
logger.info("Active profile is now '%s'%s", load_profile(pid)["name"], suffix)
return {"ok": True}
@app.post("/api/profiles/delete")
async def delete_profile(request: Request):
pid = (await request.json()).get("id", "")
if not valid_pid(pid):
return JSONResponse({"detail": "Unknown profile"}, status_code=400)
profiles = list_profiles()
if len(profiles) <= 1:
return JSONResponse({"detail": "Cannot delete the only profile"}, status_code=400)
with settings_lock:
g = load_global()
if pid == g["active_profile"] and runner and runner.is_alive():
return JSONResponse({"detail": "Stop the autopilot before deleting the active profile"}, status_code=400)
name = load_profile(pid)["name"]
shutil.rmtree(pdir(pid))
if g["active_profile"] == pid:
g["active_profile"] = next(p["id"] for p in profiles if p["id"] != pid)
save_global(g)
logger.info("Deleted profile '%s' (its videos on disk were not touched)", name)
return {"ok": True}
@app.post("/api/start")
def api_start():
global runner
if runner and runner.is_alive():
return {"ok": True, "detail": "already running"}
stop_event.clear()
runner = threading.Thread(target=run_loop, daemon=True)
runner.start()
return {"ok": True}
@app.post("/api/stop")
def api_stop():
stop_event.set()
if runner and runner.is_alive():
set_stage("stopping")
return {"ok": True}
@app.post("/api/models/pull")
async def api_pull_model(request: Request):
model = str((await request.json()).get("model", "")).strip()
if not model:
return JSONResponse({"detail": "No model name given"}, status_code=400)
g = load_global()
try:
pulled = ollama_pull(g, model)
except Exception as e:
logger.exception("Model pull failed")
return JSONResponse({"detail": str(e)}, status_code=400)
return {"ok": True, "pulled": pulled, "models": ollama_tags(g)}
@app.post("/api/history/clear")
def api_clear_history():
with state_lock:
st = load_state()
st["history"] = []
save_state(st)
logger.info("Cleared the job history")
return {"ok": True}
@app.get("/api/trending")
def api_trending():
try:
return {"items": fetch_trending(load_settings())}
except Exception as e:
logger.exception("Trending preview failed")
return JSONResponse({"detail": str(e)}, status_code=400)
@app.get("/api/logs")
def api_logs(since: int = 0):
with log_lock:
return {"lines": [x for x in log_buffer if x[0] > since]}
@app.get("/api/logs/download")
def api_logs_download():
return FileResponse(LOG_FILE, filename="app.log")
@app.get("/auth/start")
def auth_start():
global pending_flow, pending_profile
if not CLIENT_SECRET.exists():
return JSONResponse({"detail": "client_secret.json not found next to app.py"}, status_code=400)
flow = Flow.from_client_secrets_file(str(CLIENT_SECRET), scopes=SCOPES, redirect_uri=REDIRECT_URI)
url, _ = flow.authorization_url(access_type="offline", prompt="consent")
pending_flow = flow
pending_profile = load_global()["active_profile"]
return RedirectResponse(url)
@app.get("/auth/callback")
def auth_callback(request: Request):
global pending_flow, pending_profile
if pending_flow is None or not valid_pid(pending_profile):
return RedirectResponse("/")
pid = pending_profile
try:
pending_flow.fetch_token(authorization_response=str(request.url))
(pdir(pid) / "token.json").write_text(pending_flow.credentials.to_json(), encoding="utf-8")
items = youtube_client(pid).channels().list(part="snippet", mine=True).execute().get("items", [])
title = items[0]["snippet"]["title"] if items else "Linked (no channel found)"
write_json(pdir(pid) / "channel.json", {"title": title})
logger.info("Linked YouTube channel '%s' to profile '%s'", title, load_profile(pid)["name"])
except Exception as e:
logger.exception("OAuth callback failed")
return JSONResponse({"detail": str(e)}, status_code=400)
finally:
pending_flow = None
pending_profile = None
return RedirectResponse("/")
@app.post("/auth/unlink")
def auth_unlink():
pid = load_global()["active_profile"]
(pdir(pid) / "token.json").unlink(missing_ok=True)
(pdir(pid) / "channel.json").unlink(missing_ok=True)
logger.info("Unlinked YouTube channel from profile '%s'", load_profile(pid)["name"])
return {"ok": True}
if __name__ == "__main__":
import uvicorn
if not shutil.which("ffmpeg"):
logger.warning("ffmpeg not found on PATH, video stitching will fail")
uvicorn.run(app, host="127.0.0.1", port=8000)