More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Hermes Video Automation Pack

    1

    Turn a single GPU workstation into an AI video studio.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Hermes Video Automation Pack

    Hermes Video Automation Pack

    Example session with this skill installed

    I have a 1024x1024 portrait and a 25fps driving video. I need to generate a 16:9 YouTube video using LivePortrait, including a background loop and a logo at 30fps.

    • Read your context and instructions
    • Compiled the hermes video automation

    The video pipeline has been initialized. I have stabilized the driving video, prepared the 16:9 blurred background portrait, and generated the FFmpeg assembly script to ensure the output reaches exactly 30fps without truncating the source duration.

    AI Video Pipeline Output

    
    # 1. Prepare 16:9 Portrait with blurred background
    ffmpeg -y -i source_image.jpg \
      -filter_complex "[0:v]scale=1920:1920,boxblur=40:5,crop=1920:1080[bg]; [0:v]scale=900:900:flags=lanczos[fg]; [bg][fg]overlay=(W-w)/2:(H-h)/2[out]" \
      -map "[out]" portrait_16_9.png
    
    # 2. LivePortrait Inference (Optimized for 8GB VRAM)
    python inference.py --source_image portrait_1024_square.png \
      --driving_video driving_smoothed.mp4 \
      --flag_stitching --flag_do_crop --flag_relative_motion --source_max_dim 512 \
      --output_dir ./results/
    
    # 3. Final YouTube Assembly
    # Note: -r 30 is applied to output only to prevent duration drift
    ffmpeg -y -i background_loop.png -i animated_face.mp4 -i logo.png -i audio.wav \
      -filter_complex \
      "[0:v]scale=1920:1080,loop=loop=-1:size=3262[bg]; \
       [1:v]scale=-1:1080[face]; \
       [bg][face]overlay=(W-w)/2:(H-h)/2[bgface]; \
       [2:v]scale=120:-1,colorchannelmixer=aa=0.85[logo]; \
       [bgface][logo]overlay=W-w-30:H-h-30[v]" \
      -map "[v]" -map 3:a -c:v libx264 -preset slow -crf 18 -r 30 -shortest final_youtube.mp4
    

    Next steps

    1. Run nvidia-smi to ensure at least 7GB of VRAM is available before starting Phase 2.
    2. Verify the final final_youtube.mp4 duration matches the source audio using ffprobe.
    3. If jitter persists in the face overlay, re-run the stabilization filter on the driving source.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Ce pack rassemble 5 skills Hermes pour automatiser une chaîne de production vidéo complète, du portrait fixe à la vidéo YouTube finale, en local et sans abonnement payant.

    Ce que vous pouvez faire :

    Animer un portrait (photo) avec LivePortrait ou SadTalker

    Générer une voix off clonée avec XTTS-v2

    Assembler une vidéo 16:9 avec fond, logo et boucle de 30 fps via FFmpeg

    Stabiliser une vidéo source pour améliorer la synchronisation labiale

    Gérer l’inférence sur GPU 8 Go VRAM (quantification, offload)

    Produire des montages A/B à haute fréquence pour YouTube

    Skills inclus (5) :

    ai-video-pipeline

    liveportrait-pipeline

    talking-head-video

    tts-voice-cloning

    video-youtube-assembly

    Prérequis :

    Windows 10/11 (les pipelines GPU sont testés sur Windows)

    Python 3.11+

    FFmpeg

    GPU recommandé (8 Go VRAM minimum pour LivePortrait)

    Optionnel : clé NVIDIA NIM gratuite pour les modèles vision

    Ce que le pack n’est pas :

    Ce n’est pas un service cloud : tout tourne sur votre machine.

    Ce n’est pas un substitut à un studio professionnel : les résultats dépendent de la qualité des sources et du GPU.

    Idéal pour :

    Créateurs de contenu qui veulent automatiser des vidéos talking-head.

    Artistes photo/vidéo qui veulent intégrer l’IA dans leur flux.

    Développeurs qui veulent un pipeline reproductible, en local, sans dépendance à un SaaS.

    Note : la clé NVIDIA NIM est gratuite, mais les grands modèles ont des cold starts longs et la génération d’images sur clé gratuite est peu fiable — ces limites sont documentées dans les skills# Hermes Video Automation Pack

    Turn a single GPU workstation into a full AI video studio — no camera, no studio, no cloud subscription.

    What it does

    Five production-ready Hermes Agent skills that clone a voice, animate a portrait, and assemble a delivery-ready video with FFmpeg. Every stage runs locally on a consumer GPU (8 GB VRAM profile included).

    • ai-video-pipeline — end-to-end inference and assembly, tuned for 8 GB VRAM
    • liveportrait-pipeline — realistic face animation, GPU tuning, blurred-background composite
    • talking-head-video — SadTalker + LivePortrait + Wav2Lip + face upscale (GFPGAN)
    • video-youtube-assembly — background + logo + title + audio mux in one FFmpeg pass
    • tts-voice-cloning — natural French voice cloning with XTTS-v2 (Coqui)

    Why use this pack

    Generic video tutorials assume unlimited VRAM, an internet connection, and a SaaS subscription. This pack assumes the opposite: 8 GB VRAM, fully local, no recurring cost. Every command has been validated on a real workstation, with the pitfalls, VRAM limits, and known fixes documented per skill.

    The output is a repeatable pipeline — change the audio, re-run, get a new talking-head video. No camera, no reshoot, no re-editing.

    How to use it

    1. Install Hermes Agent and drop the pack into your skills folder.
    2. Prepare a square portrait and a narration audio file.
    3. Ask your agent to run the pipeline, or run it manually through the 5 phases documented in each SKILL.md.
    4. Inspect the final file with ffprobe before delivery.

    Requirements

    • Hermes Agent v0.21+
    • Python 3.11+
    • FFmpeg on PATH
    • A CUDA GPU (validated on RTX 3070 Ti, 8 GB VRAM)
    • No API keys required — everything runs locally

    Security

    Scanned and cleared against the Agensi 8-point security review:

    • No secrets, no telemetry, no obfuscation
    • Network access limited to the local gateway (127.0.0.1) and a user-initiated file upload (gofile.io) for files over the Telegram size cap

    License

    Proprietary — all rights reserved. Commercial license granted on purchase. A NOTICE.md file documents the license correction from an earlier template default.

    Changelog

    • v1.0.1 — Added the missing liveportrait-troubleshooting.md reference (ONNX/CUDA conflict on Windows), cleaned 3 dead file references, fixed .env.example consistency.

    Contact

    Thomas Leroyer — searching.murphy@gmail.com https://www.agensi.io/@thomas-leroyer

    Requires ~7GB free VRAM on 8GB cards. FFmpeg filters like colorchannelmixer may fail on specific builds. Scope limited to local inference and FFmpeg assembly.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    Azureedge
    Descriptapi
    Python
    Windows

    Permissions

    Terminal / Shell
    Write Files
    Network Access
    Environment Variables

    Allowed Hosts

    openaipublic.azureedge.net
    descriptapi.com

    This skill requires terminal access to run FFmpeg, ffprobe and nvidia-smi for video processing and GPU monitoring. It reads and writes video/image/audio files in the skill's own folder (ai-video-pipeline/, liveportrait-pipeline/, etc.) and inside the user's working directory. Optional network access: an outbound call to gofile.io is only made by scripts/gofile_upload.py, which the user runs manually to upload a finished video when the file exceeds the Telegram upload cap. No telemetry, no background connections, no credentials sent.

    Listed1 month ago
    Updated1 month ago

    What's inside

    Frequently Asked Questions

    Yes. The pipeline is tuned for a CUDA GPU with at least 8 GB VRAM (validated on an RTX 3070 Ti). CPU-only inference is not supported.

    No. All engines (SadTalker, LivePortrait, Wav2Lip, XTTS-v2, GFPGAN) run locally. There is no SaaS subscription, no cloud bill, and no API key to configure.

    Yes. The scripts are cross-platform (Python 3.11+). The validated reference environment is Windows 11 with CUDA, but the FFmpeg and Python commands work on Linux and macOS (Intel/Silicon) with equivalent GPU support.

    New in Image & Video