convez.ai
SkillAutomaçãoAvançado
manvendrakumar-hub

Product Modelling — Cloning de Influenciador

manvendrakumar-hub/product-modelling-cloning-influencer

Pipeline zero-install para product modelling com avatar de referência real

Ver no GitHub
InicialPluginScore · de 0 a 100

Pipeline reutilizável (Claude skill + scripts Python) para clonar um modelo a partir de vídeo de referência e produzir vídeos de product modelling (óculos, roupas, acessórios) sem distorcer o rosto. Inclui ferramentas para extração de frames, geração de character sheets, virtual try-on e montagem com ffmpeg — traz apenas a metodologia e tooling, você fornece material com direitos/consentimento.

Instalação

$npx convez-ai manvendrakumar-hub/product-modelling-cloning-influencer

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerar vídeos de produto (eyewear, apparel, accessories) com modelo consistente
  • Criar posing videos e virtual try-on sem distorção facial
  • Montar reels de fast-cut com efeitos locais via ffmpeg

Veredito

Skill sofisticada e bem documentada para automação de video production em product modelling. Requer material próprio com direitos garantidos e conhecimento técnico; não é plug-and-play, mas oferece pipeline robusto e metodologia clara baseada em boas práticas éticas.

README

A tradução para o português deste README ainda está na fila.

Product Modelling — Cloning an Influencer

A reusable, zero-install pipeline (packaged as a Claude Code / Agent skill) for cloning a real model or influencer from a reference video and producing product-modelling videos — eyewear, apparel, accessories — without distorting the person's face.

⚠️ This repo ships the method and tooling only. It contains no real-person likeness media, no cloned faces, no brand footage, and no product photography. Bring your own properly-licensed source material, and make sure you have the rights/consent to clone any real person before you do.

What's inside

SKILL.md                         # the full 8-step pipeline + operating principles
references/
  engine-catalog.md              # which generation model for which job (image + video)
  gotchas-and-uploads.md         # upload flow, large-file PUT fix, content-filter & preset gotchas
  posing-prompt-library.md       # researched fashion posing + hand-gesture prompt phrases
  character-sheet-template.md    # the written face-spec format + face-lock prompt fragment
scripts/
  build_sheet.py                 # build a labelled character/reference sheet from real frames (PIL)
  build_director_grid.py         # build a creative-director storyboard grid sheet
  build_reel.py                  # local fast-cut montage (ffmpeg trim/crop/concat, punch-in zooms)

The pipeline (TL;DR)

  1. Extract frames from the reference clip (bundled ffmpeg via imageio_ffmpeg — no system ffmpeg).
  2. Pick the hero frame (clean frontal / three-quarter / profile).
  3. Build a character sheet — written face spec + a labelled montage from the real frames.
  4. Generate identity-locked stills — face sheet, full-body turnaround, wardrobe/product variants.
  5. Virtual try-on / product ad — avatar + product (wardrobe is driven by the avatar image).
  6. Cinematic shots — image-to-video per storyboard beat, then stitch.
  7. Posing videos — best on a posing-capable engine; use the posing-phrase library.
  8. Fast-cut montage — local ffmpeg edit; QA every output frame-by-frame.

The golden rule

Do not distort the face. Anchor every generation to the real reference frames, and prefer real-frame edits when you can avoid AI-inventing a face at all.

Fast-cut montage (Step 8)

build_reel.py hard-cuts clean takes into a punchy reel with punch-in "zoom" reframes — locally, zero credits. Synthetic, likeness-free preview of the style:

The character sheet

Step 3 produces a labelled reference sheet (built from real frames) plus a written face spec that becomes the prompt backbone for every later generation. Layout (placeholder silhouettes shown — no real likeness ships in this repo):

Requirements

  • Python 3 with imageio, imageio-ffmpeg, Pillow (pip3 install --user imageio imageio-ffmpeg pillow).
  • An image/video generation backend of your choice (the docs reference the models used in the original build).

Credits & inspiration

QA loop adapted from a grid-review approach (frame-by-frame contact sheets); prompt/camera craft informed by a Seedance director prompt system. Engine-specific notes reflect one real production run.

Responsible use

Only clone real people with their consent. Respect third-party copyright (brand films, product imagery) and right of publicity / privacy. Don't publish or distribute a person's AI likeness without permission. This tooling is provided for legitimate, authorized creative work.

Ver o original no GitHub

Bom para

Relacionados