luke/ai-podcast
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Make make_clips.py resilient — timeouts, retries, skip-on-failure

Browse Source 
- 60s timeout + retry on all LLM calls
- 120-300s timeout on all subprocess/ffmpeg calls
- Per-clip error isolation (one failure doesn't kill the run)
- Progress indicators for each clip being processed

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
tcpsyn
2026-03-20 17:36:41 -06:00
parent 4589670b37
commit e0fb3cac68
1 changed files with 142 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files
make_clips.py
+142 -114
View File
@@ -23,6 +23,8 @@ import tempfile
import xml.etree.ElementTree as ET import xml.etree.ElementTree as ET
from pathlib import Path from pathlib import Path
import time
import requests import requests
from dotenv import load_dotenv from dotenv import load_dotenv
@@ -46,6 +48,50 @@ WIDTH = 1080
HEIGHT = 1920 HEIGHT = 1920
def _llm_request(prompt: str, max_tokens: int = 2048, temperature: float = 0.3,
timeout: int = 60) -> str | None:
"""Make an LLM API call with timeout and retry. Returns content or None on failure."""
for attempt in range(2):
try:
response = requests.post(
"https://openrouter.ai/api/v1/chat/completions",
headers={
"Authorization": f"Bearer {OPENROUTER_API_KEY}",
"Content-Type": "application/json",
},
json={
"model": "anthropic/claude-sonnet-4-5",
"messages": [{"role": "user", "content": prompt}],
"max_tokens": max_tokens,
"temperature": temperature,
},
timeout=timeout,
)
if response.status_code != 200:
print(f" LLM error (HTTP {response.status_code}): {response.text[:200]}")
if attempt == 0:
print(f" Retrying in 5s...")
time.sleep(5)
continue
return None
return response.json()["choices"][0]["message"]["content"].strip()
except requests.Timeout:
print(f" LLM request timed out ({timeout}s)")
if attempt == 0:
print(f" Retrying in 5s...")
time.sleep(5)
continue
return None
except Exception as e:
print(f" LLM request failed: {e}")
if attempt == 0:
print(f" Retrying in 5s...")
time.sleep(5)
continue
return None
return None
def _build_whisper_prompt(labeled_transcript: str) -> str: def _build_whisper_prompt(labeled_transcript: str) -> str:
"""Build an initial_prompt for Whisper from the labeled transcript. """Build an initial_prompt for Whisper from the labeled transcript.
@@ -186,7 +232,12 @@ def refine_clip_timestamps(audio_path: str, clips: list[dict],
"ffmpeg", "-y", "-ss", str(seg_start), "-t", str(seg_end - seg_start), "ffmpeg", "-y", "-ss", str(seg_start), "-t", str(seg_end - seg_start),
"-i", audio_path, "-ar", "16000", "-ac", "1", seg_path, "-i", audio_path, "-ar", "16000", "-ac", "1", seg_path,
] ]
result = subprocess.run(cmd, capture_output=True, text=True) try:
result = subprocess.run(cmd, capture_output=True, text=True, timeout=120)
except subprocess.TimeoutExpired:
print(f" Clip {i+1}: ffmpeg timed out (120s), skipping")
refined[i] = []
continue
if result.returncode != 0: if result.returncode != 0:
print(f" Clip {i+1}: Failed to extract segment") print(f" Clip {i+1}: Failed to extract segment")
refined[i] = [] refined[i] = []
@@ -279,25 +330,11 @@ IMPORTANT:
Respond with ONLY a JSON array, no markdown or explanation: Respond with ONLY a JSON array, no markdown or explanation:
[{{"title": "...", "start_time": 0.0, "end_time": 0.0, "caption_text": "..."}}]""" [{{"title": "...", "start_time": 0.0, "end_time": 0.0, "caption_text": "..."}}]"""
response = requests.post( content = _llm_request(prompt, max_tokens=2048, temperature=0.3, timeout=60)
"https://openrouter.ai/api/v1/chat/completions", if content is None:
headers={ print(" Failed to get clip selections from LLM — aborting")
"Authorization": f"Bearer {OPENROUTER_API_KEY}", return []
"Content-Type": "application/json",
},
json={
"model": "anthropic/claude-sonnet-4-5",
"messages": [{"role": "user", "content": prompt}],
"max_tokens": 2048,
"temperature": 0.3,
},
)
if response.status_code != 200:
print(f"Error from OpenRouter: {response.text}")
sys.exit(1)
content = response.json()["choices"][0]["message"]["content"].strip()
if content.startswith("```"): if content.startswith("```"):
content = re.sub(r"^```(?:json)?\n?", "", content) content = re.sub(r"^```(?:json)?\n?", "", content)
content = re.sub(r"\n?```$", "", content) content = re.sub(r"\n?```$", "", content)
@@ -307,7 +344,7 @@ Respond with ONLY a JSON array, no markdown or explanation:
except json.JSONDecodeError as e: except json.JSONDecodeError as e:
print(f"Error parsing LLM response: {e}") print(f"Error parsing LLM response: {e}")
print(f"Response was: {content[:500]}") print(f"Response was: {content[:500]}")
sys.exit(1) return []
# Validate and clamp durations # Validate and clamp durations
validated = [] validated = []
@@ -349,25 +386,11 @@ For each clip, generate:
Respond with ONLY a JSON array matching the clip order: Respond with ONLY a JSON array matching the clip order:
[{{"description": "...", "hashtags": ["#tag1", "#tag2", ...]}}]""" [{{"description": "...", "hashtags": ["#tag1", "#tag2", ...]}}]"""
response = requests.post( content = _llm_request(prompt, max_tokens=2048, temperature=0.7, timeout=60)
"https://openrouter.ai/api/v1/chat/completions", if content is None:
headers={ print(" Failed to generate social metadata — skipping")
"Authorization": f"Bearer {OPENROUTER_API_KEY}",
"Content-Type": "application/json",
},
json={
"model": "anthropic/claude-sonnet-4-5",
"messages": [{"role": "user", "content": prompt}],
"max_tokens": 2048,
"temperature": 0.7,
},
)
if response.status_code != 200:
print(f"Error from OpenRouter: {response.text}")
return clips return clips
content = response.json()["choices"][0]["message"]["content"].strip()
if content.startswith("```"): if content.startswith("```"):
content = re.sub(r"^```(?:json)?\n?", "", content) content = re.sub(r"^```(?:json)?\n?", "", content)
content = re.sub(r"\n?```$", "", content) content = re.sub(r"\n?```$", "", content)
@@ -777,43 +800,25 @@ RULES:
RAW TEXT ({len(words)} words): RAW TEXT ({len(words)} words):
{raw_text}""" {raw_text}"""
try: polished = _llm_request(prompt, max_tokens=2048, temperature=0, timeout=30)
response = requests.post( if polished is None:
"https://openrouter.ai/api/v1/chat/completions", print(f" Polish failed, using raw text")
headers={ return words
"Authorization": f"Bearer {OPENROUTER_API_KEY}",
"Content-Type": "application/json",
},
json={
"model": "anthropic/claude-sonnet-4-5",
"messages": [{"role": "user", "content": prompt}],
"max_tokens": 2048,
"temperature": 0,
},
timeout=30,
)
if response.status_code != 200:
print(f" Polish failed ({response.status_code}), using raw text")
return words
polished = response.json()["choices"][0]["message"]["content"].strip() polished_words = polished.split()
polished_words = polished.split()
if len(polished_words) != len(words): if len(polished_words) != len(words):
print(f" Polish word count mismatch ({len(polished_words)} vs {len(words)}), using raw text") print(f" Polish word count mismatch ({len(polished_words)} vs {len(words)}), using raw text")
return words return words
changes = 0 changes = 0
for i, pw in enumerate(polished_words): for i, pw in enumerate(polished_words):
if pw != words[i]["word"]: if pw != words[i]["word"]:
changes += 1 changes += 1
words[i]["word"] = pw words[i]["word"] = pw
if changes: if changes:
print(f" Polished {changes} words") print(f" Polished {changes} words")
except Exception as e:
print(f" Polish error: {e}")
return words return words
@@ -898,8 +903,12 @@ def extract_clip_audio(audio_path: str, start: float, end: float,
output_path, output_path,
] ]
result = subprocess.run(cmd, capture_output=True, text=True) try:
return result.returncode == 0 result = subprocess.run(cmd, capture_output=True, text=True, timeout=120)
return result.returncode == 0
except subprocess.TimeoutExpired:
print(f" ffmpeg audio extraction timed out (120s)")
return False
def generate_background_image(episode_number: int, clip_title: str, def generate_background_image(episode_number: int, clip_title: str,
@@ -1153,7 +1162,11 @@ def generate_clip_video(audio_path: str, background_path: str,
output_path, output_path,
] ]
result = subprocess.run(cmd, capture_output=True, text=True) try:
result = subprocess.run(cmd, capture_output=True, text=True, timeout=300)
except subprocess.TimeoutExpired:
print(f" ffmpeg video generation timed out (300s)")
return False
if result.returncode != 0: if result.returncode != 0:
print(f" ffmpeg error: {result.stderr[-300:]}") print(f" ffmpeg error: {result.stderr[-300:]}")
return False return False
@@ -1235,7 +1248,12 @@ def generate_clip_video_remotion(
output_path, output_path,
] ]
result = subprocess.run(cmd, capture_output=True, text=True, cwd=str(REMOTION_DIR)) try:
result = subprocess.run(cmd, capture_output=True, text=True, cwd=str(REMOTION_DIR), timeout=180)
except subprocess.TimeoutExpired:
props_path.unlink(missing_ok=True)
print(f" Remotion render timed out (180s)")
return False
props_path.unlink(missing_ok=True) props_path.unlink(missing_ok=True)
if result.returncode != 0: if result.returncode != 0:
@@ -1488,6 +1506,9 @@ def main():
print(f"\n[3/{step_total}] Selecting {args.count} best moments with LLM...") print(f"\n[3/{step_total}] Selecting {args.count} best moments with LLM...")
clips = select_clips_with_llm(transcript_text, labeled_transcript, clips = select_clips_with_llm(transcript_text, labeled_transcript,
chapters_json, args.count) chapters_json, args.count)
if not clips:
print("\nNo clips selected — aborting.")
return
# Snap to sentence boundaries so clips don't start/end mid-sentence # Snap to sentence boundaries so clips don't start/end mid-sentence
clips = snap_to_sentences(clips, segments) clips = snap_to_sentences(clips, segments)
@@ -1524,14 +1545,18 @@ def main():
extract_step = 6 if two_pass else 5 extract_step = 6 if two_pass else 5
print(f"\n[{extract_step}/{step_total}] Extracting audio clips...") print(f"\n[{extract_step}/{step_total}] Extracting audio clips...")
for i, clip in enumerate(clips): for i, clip in enumerate(clips):
print(f" [{i+1}/{len(clips)}] \"{clip['title']}\"...")
slug = slugify(clip["title"]) slug = slugify(clip["title"])
mp3_path = output_dir / f"clip-{i+1}-{slug}.mp3" mp3_path = output_dir / f"clip-{i+1}-{slug}.mp3"
if extract_clip_audio(str(audio_path), clip["start_time"], clip["end_time"], try:
str(mp3_path)): if extract_clip_audio(str(audio_path), clip["start_time"], clip["end_time"],
print(f" Clip {i+1} audio: {mp3_path.name}") str(mp3_path)):
else: print(f" Clip {i+1} audio: {mp3_path.name}")
print(f" Error extracting clip {i+1} audio") else:
print(f" Error extracting clip {i+1} audio — skipping")
except Exception as e:
print(f" Clip {i+1} audio failed: {e} — skipping")
video_step = 7 if two_pass else 6 video_step = 7 if two_pass else 6
if args.audio_only: if args.audio_only:
@@ -1553,49 +1578,52 @@ def main():
mp4_path = output_dir / f"clip-{i+1}-{slug}.mp4" mp4_path = output_dir / f"clip-{i+1}-{slug}.mp4"
duration = clip["end_time"] - clip["start_time"] duration = clip["end_time"] - clip["start_time"]
print(f" Clip {i+1}: Generating video...") print(f" [{i+1}/{len(clips)}] \"{clip['title']}\" ({duration:.0f}s)...")
# Get word timestamps — use refined segments if available try:
word_source = refined[i] if (two_pass and i in refined and refined[i]) else segments # Get word timestamps — use refined segments if available
clip_words = get_words_in_range(word_source, clip["start_time"], clip["end_time"]) word_source = refined[i] if (two_pass and i in refined and refined[i]) else segments
clip_words = get_words_in_range(word_source, clip["start_time"], clip["end_time"])
# Add speaker labels # Add speaker labels
clip_words = add_speaker_labels(clip_words, labeled_transcript, clip_words = add_speaker_labels(clip_words, labeled_transcript,
clip["start_time"], clip["end_time"], clip["start_time"], clip["end_time"],
word_source) word_source)
# Polish text with LLM (fix punctuation, capitalization, mishearings) # Polish text with LLM (fix punctuation, capitalization, mishearings)
clip_words = polish_clip_words(clip_words, labeled_transcript) clip_words = polish_clip_words(clip_words, labeled_transcript)
# Group words into timed caption lines # Group words into timed caption lines
caption_lines = group_words_into_lines( caption_lines = group_words_into_lines(
clip_words, clip["start_time"], duration clip_words, clip["start_time"], duration
) )
if use_remotion: if use_remotion:
if generate_clip_video_remotion( if generate_clip_video_remotion(
str(mp3_path), caption_lines, clip["start_time"], str(mp3_path), caption_lines, clip["start_time"],
clip["title"], episode_number, str(mp4_path), duration clip["title"], episode_number, str(mp4_path), duration
): ):
file_size = mp4_path.stat().st_size / (1024 * 1024) file_size = mp4_path.stat().st_size / (1024 * 1024)
print(f" Clip {i+1} video: {mp4_path.name} ({file_size:.1f} MB)") print(f" Clip {i+1} video: {mp4_path.name} ({file_size:.1f} MB)")
else:
print(f" Clip {i+1} video failed (Remotion) — skipping")
else: else:
print(f" Error generating clip {i+1} video (Remotion)") # Legacy PIL+ffmpeg renderer
else: bg_path = str(tmp_dir / f"bg_{i}.png")
# Legacy PIL+ffmpeg renderer generate_background_image(episode_number, clip["title"], bg_path)
bg_path = str(tmp_dir / f"bg_{i}.png")
generate_background_image(episode_number, clip["title"], bg_path)
clip_tmp = tmp_dir / f"clip_{i}" clip_tmp = tmp_dir / f"clip_{i}"
clip_tmp.mkdir(exist_ok=True) clip_tmp.mkdir(exist_ok=True)
if generate_clip_video(str(mp3_path), bg_path, caption_lines, if generate_clip_video(str(mp3_path), bg_path, caption_lines,
clip["start_time"], str(mp4_path), clip["start_time"], str(mp4_path),
duration, clip_tmp): duration, clip_tmp):
file_size = mp4_path.stat().st_size / (1024 * 1024) file_size = mp4_path.stat().st_size / (1024 * 1024)
print(f" Clip {i+1} video: {mp4_path.name} ({file_size:.1f} MB)") print(f" Clip {i+1} video: {mp4_path.name} ({file_size:.1f} MB)")
else: else:
print(f" Error generating clip {i+1} video") print(f" Clip {i+1} video failed (ffmpeg) — skipping")
except Exception as e:
print(f" Clip {i+1} video failed: {e} — skipping")
# Save clips metadata for social upload # Save clips metadata for social upload
metadata_path = output_dir / "clips-metadata.json" metadata_path = output_dir / "clips-metadata.json"