JPeetz/agent-skills / skills/marketing/agent-n-video-pipeline
agent-n-video-pipeline
UnverifiedUse this skill when producing a fully-local "Agent N presents a message" video on the 16GB Apple-Silicon M5 executor: Kokoro mature TTS → LTX-2.3 MLX A2V talking render with native lip sync → karaoke captions → trim → QA → X caption. Replaces the old cloud Herald pipeline (grok-imagine + volcengine) with zero cloud cost. Activates for Herald release videos, Agent N announcements, and newsletter video messages.
| File | Kind | Size |
|---|---|---|
| CHANGELOG.md | markdown | 523 B |
| SKILL.md | skill md | 4.4 KB |
| evals/evals.json | config | 2.4 KB |
| scripts/build_final_video_m5.py | script | 5.0 KB |
| scripts/herald_pipeline_m5.shexecutable | script | 5.4 KB |
| scripts/karaoke_burn.py | script | 11.8 KB |
| scripts/kokoro_tts.py | script | 2.2 KB |
scripts/build_final_video_m5.py
View on GitHub#!/usr/bin/env python3
"""Build a Hermes Herald video — FULLY LOCAL on the M5 executor.
Translation of build_final_video.py (grok-imagine + volcengine cloud legs) to
local models: LTX-2.3 MLX A2V (talking render + native lip-sync in one leg),
Piper TTS (mature voice), then the SAME local caption/trim/QA stages.
Usage:
python3 build_final_video_m5.py --script "<speech>" [--out NAME] [--duration 22] [--voice PATH] [--anchor PATH]
or --spec spec.json (same schema as the cloud version: script/prompt/out/duration/voice)
Stages:
1. TTS (Piper, mature voice) -> audio.wav
2. LTX-2.3 MLX A2V on M5 (anchor + audio -> talking video with lip sync)
3. karaoke_burn.py (same as cloud version, PressStart2P word-synced)
4. trim to speech+2.5s
5. qa_review.py (same deterministic gate)
6. x_caption.py
"""
import re, json, time, os, sys, subprocess, argparse, urllib.request
AP = argparse.ArgumentParser()
AP.add_argument("--script", default=None)
AP.add_argument("--spec", default=None)
AP.add_argument("--out", default="hermes-herald-m5-FINAL.mp4")
AP.add_argument("--duration", type=int, default=22)
AP.add_argument("--voice", default="/tmp/m5_images/audio/hn_mature.wav")
AP.add_argument("--anchor", default="agentn_anchor_v1.png")
AP.add_argument("--m5", default="192.168.212.46")
args = AP.parse_args()
BASE = "/Users/joergpeetz/workspace/scripts/hermes_build"
M5 = f"http://{args.m5}:8188"
WORK = "/tmp/m5_herald"
os.makedirs(WORK, exist_ok=True)
KARAOKE = "/Users/joergpeetz/workspace/scripts/karaoke_burn.py"
KOKORO_PY = "/Users/joergpeetz/.venvs/kokoro/bin/python"
PY = "/usr/local/bin/python3"
# ---------------------------------------------------------------- spec resolve
if args.spec:
spec = json.load(open(args.spec))
SCRIPT = spec["script"]
args.out = spec.get("out", args.out)
args.duration = int(spec.get("duration", args.duration))
if spec.get("voice"):
args.voice = spec.get("voice")
else:
SCRIPT = args.script or ("Hermes researches with receipts now. Every claim is matched to a "
"real source, cited, and fact-checked before it reaches you. Hand it any document "
"\u2014 it shows what checks out, and what doesn't. That's trust you can see.")
# ---------------------------------------------------------------- 1. TTS (Kokoro sid 10 = mature voice)
print("── 1/6 TTS (Kokoro sid 10, mature) ──")
tts_py = "/Users/joergpeetz/workspace/scripts/kokoro_tts.py"
tts_out = os.path.join(WORK, "voice.wav")
if os.path.exists(args.voice) and os.path.getsize(args.voice) > 10000:
subprocess.run(["cp", args.voice, tts_out], check=True)
print(f" using supplied voice: {args.voice}")
else:
subprocess.run([KOKORO_PY, tts_py, "10", "1.0", tts_out, SCRIPT], check=True, timeout=300)
print(f" Kokoro synth done: {tts_out}")
print(f" voice: {tts_out} ({os.path.getsize(tts_out)} bytes)")
# ---------------------------------------------------------------- 2. LTX A2V on M5
print("── 2/6 LTX-2.3 MLX A2V on M5 ──")
# stage anchor + audio into M5 ComfyUI input dir
subprocess.run(["sshpass", "-p", "Buddy-2019", "scp", "-o", "StrictHostKeyChecking=no",
os.path.join(WORK, "voice.wav"),
f"joergpeetz@{args.m5}:/Users/joergpeetz/ComfyUI-Shared/input/agentn_speech.wav"], check=True)
a2v_prompt = (
"Anime girl with short black bob hair and black over-ear headphones with orange accents "
"speaks directly to the camera, natural facial expressions and lip movement, subtle "
"head movement, gentle confident tone, semi-realistic anime illustration, sharp focus, "
"clean dark studio backdrop with violet grade."
)
frames = int(args.duration * 24)
wf = {
"img": {"class_type": "LoadImage", "inputs": {"image": args.anchor}},
"aud": {"class_type": "LoadAudio", "inputs": {"audio": "agentn_speech.wav"}},
"a2v": {"class_type": "LTXMLXA2V", "inputs": {
"image": ["img", 0], "audio": ["aud", 0], "prompt": a2v_prompt,
"width": 512, "height": 768, "frames": min(frames, 200), "frame_rate": 24,
"seed": 42, "cfg_scale": 3.0, "low_ram": True,
"model_dir": "dgrauet/ltx-2.3-mlx-q8"}},
"video": {"class_type": "CreateVideo", "inputs": {
"images": ["a2v", 0], "fps": 24.0, "audio": ["a2v", 1]}},
"save": {"class_type": "SaveVideo", "inputs": {
"video": ["video", 0], "filename_prefix": "agentn_herald_a2v", "format": "mp4", "codec": "h264"}},
}
r = subprocess.run(["sshpass", "-p", "Buddy-2019", "ssh", "-o", "StrictHostKeyChecking=no",
f"joergpeetz@{args.m5}",
f"curl -s --max-time 20 -X POST http://127.0.0.1:8188/prompt -H 'Content-Type: application/json' -d @/tmp/m5_herald_wf.json"],
capture_output=True, text=True)
print(" queued response:", r.stdout.strip()[:120])
# ---------------------------------------------------------------- 3-6 local stages (unchanged from cloud pipeline)
print("─ local stages 3-6 (karaoke/trim/QA/caption) run in full pipeline after A2V output is pulled ─")Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |