Apresentadora Virtual Realista · Open Skill 公开技能
Apresentadora Virtual Realista
A Apresentadora Realista AI transforma uma apresentadora de referência e um roteiro em uma produção audiovisual completa. A Skill analisa o comportamento da apresentadora — postura, gestos, olhar, expressões, ritmo corporal, movimentos das mãos, cabeça e microexpressões — e transforma essas características em um perfil de comportamento consistente. Também cria e dirige o ambiente da cena, considerando arquitetura, profundidade, iluminação, clima, elementos em movimento e interação da apresentadora com o espaço. A cinematografia é planejada de forma intencional, utilizando diferentes lentes, profundidade de campo, desfoque, enquadramentos, ângulos e movimentos de câmera como dolly, tilt, pan, orbit, handheld, POV, FPV, zoom e outros movimentos cinematográficos. A Skill também desenvolve o design de som da cena, incluindo voz, ambiente, Foley e efeitos sonoros, podendo incorporar sons urbanos, natureza, pássaros, vento, água, máquinas e outros elementos do ambiente. O roteiro também orienta a performance: intenção, emoção, energia, olhar, gestos, pausas e relação entre atuação e câmera. O resultado é uma apresentadora integrada ao ambiente, à câmera, à iluminação, ao áudio e à edição — e não simplesmente uma pessoa falando diante de um fundo.
Explainer
23 3 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/2160607/2026-09-24/rgzn2m7r.png
# Apresentadora Virtual Realista
DIGEN skill that turns an audio or text brief into a broadcast-realistic talking-head presenter video, for an agency's own services or for a client in any niche (law, medicine, aesthetics, real estate, education, restaurants, corporate). There is no structured project state; progress lives in the conversation, canvas assets, and `write_todos`.
## Global Language Lock (mandatory, highest priority)
This Skill and every file under `references/` are **English only** (handbook). Handbook English is never the reply language by itself — detect and lock the reply language from the user's own typed words on the first real message, ignoring any platform "invoked via /" heading and any `<untrusted_skill_content>` wrapper around this Skill.
| Layer | Scope | Language |
|---|---|---|
| `response_language` | Everything the user reads: replies, tables, buttons, todos, canvas document titles/bodies (except spoken-line quotes) | Locked on turn 1; one language per turn, never mixed |
| `narration_language` | The presenter's spoken lines: script text, TTS `text`, dialogue shown in a shot table | Defaults to the client's own language for this project (commonly `pt-BR` for a Brazilian client); changes only on an explicit request |
| Model prompts | `generate_image` / `generate_video` / `generate_music` prompt bodies, `voice_instruction` | **Always English** |
Never narrate internal steps ("reading the skill", "checking the stage", "locking language") in any language. The first visible sentence is about the brief itself — niche, presenter, scene, lines, or the cut — never about protocol.
## When to use
- An agency wants a presenter video for its own services, or for a client in any niche.
- The user provides an audio brief, a script, or just a topic and expects a finished talking-head video (**MODE A — Creation**).
- The user provides ready-made final image(s) + final audio + script + a scene sequence and wants the material **animated, not recreated** (**MODE B — Preservation/Animation**).
- The user wants to revise one stage only (persona, scene, lines, voice, one clip), or produce a 9:16 cut from an existing 16:9 project.
## Two modes (decide on the first message)
**Detect the mode in intake — it is the first decision and the two modes never mix.**
- **MODE A — CREATION:** the user gives a theme/brief ("create a presenter on a farm talking about agribusiness"). The skill defines persona, scene, camera, lighting, behavior, performance, generates the image, generates TTS, and renders the video. Full generation pipeline.
- **MODE B — PRESERVATION / ANIMATION:** the user supplies final image(s) + final audio + script + a scene sequence and asks to "animate / preserve / use this material". The skill does **not** recreate anything — it animates the existing material.
**Detection rule:** if the user attaches final image(s) **and** a final audio and asks to animate/preserve/use them → **MODE B**. If the user gives only a topic/brief → **MODE A**. When in doubt, ask. Never apply Mode A generation rules to Mode B material, and never apply Mode B preservation rules to a Mode A brief.
### MODE B — core rules (see `references/10-preservation-mode.md`)
1. **VISUAL LOCK:** the user's image is the final composition. Preserve identity, face, hair, age, clothing, body, body position, vehicle, interior, scene, objects, existing lighting, perspective, camera position/distance, framing, character scale, foreground/midground/background relationship, visual direction, and overall photographic look.
2. **Never describe** the character, clothing, scene, camera, lens, lighting or environment in the prompt when a ready image exists — reference `[@ImageN]` only.
3. **CAMERA LOCKED:** no zoom, crop, reframe, push-in, dolly, handheld sway, lens change, perspective change, or recomposition. The image's framing is the scene's framing.
4. **ANIMATE WITHOUT REINTERPRETING:** only mouth movement, blinking, eye micro-motion, micro-expre