Become a Vlogger · skill pública de Open Skill

Become a Vlogger

Become a Vlogger es una Open Skill pública en Digen. Ábrela para ver qué crea este flujo.

@jeeai

VTuber

10 3 6

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1821775/2026-09-24/2hxhar3k.jpg

# Hyper-Realistic Vlog Video Creator

## When to use
- The user wants a realistic, natural, cinematic vlog video for YouTube / Instagram / Facebook / other social platforms, featuring a host, creator, family, traveller, student, entrepreneur, chef, artist, farmer, fitness enthusiast, or an authorised reference-image person.
- The user provides an idea, a script, a reference image, an uploaded video, or an uploaded audio track and wants it turned into a vlog.
- The user wants to revise or continue an existing vlog package (change the script, restyle a scene, redo a clip, generate the thumbnail/voiceover/music, or produce the final video).

## Reference files
- `references/production-spec.md`: the full vlog production rules — vlog types, default production, reference-person rules, host profile, identity/continuity, story structure, script style, reference-media handling, hyper-realistic visual style, camera modes/shots, lighting, audio, B-roll, facial expressions, gestures, timeline format, and the required-output template. Read this file before producing any package.

## Steps
1. **Clarify** (one round): vlog type, topic/place/activity, duration (default 30–60s if not given), platform, language, aspect ratio (16:9 landscape unless vertical requested), and whether a reference image / uploaded video / uploaded audio / script is provided.
2. **Ingest reference media**:
   - Reference image → `recognize_image(image_url=..., prompt="Describe the person's face, skin tone, hairstyle, age, body proportions, clothing, and the setting so they can be kept consistent.")`.
   - Uploaded video → `recognize_video(video_url=..., prompt="Describe the action, setting, shots, and visual style.")` and, if dialogue is needed, `transcribe_audio(audio_url=...)`.
   - Uploaded audio → `transcribe_audio(audio_url=...)` to get the exact timing source (dialogue, pauses, breaths).
   - Script → use it as the narration basis.
3. **Build the production package** with `create_document(content=..., name="<vlog title> — production package", content_type="text/markdown", entity_type="scripts")`, following the required-output template in `references/production-spec.md` (title, type, duration, aspect ratio, resolution, target audience, language, location, host, opening hook, full script, host profile, location/world design, shot list, B-roll plan, camera plan, lighting plan, audio plan, editing plan, thumbnail text/prompt, YouTube title/description, short-form caption, keywords, hashtags, call to action, disclosure). Confirm the package with the user before generating media.
4. **Generate host / scene stills** with `generate_image`:
   - No reference image → `model='t2i'`; describe the host and scene per the hyper-realistic style rules.
   - Reference image provided → `model='i2i'` with `input_urls=[<reference asset>]` to keep the person consistent.
   - Use `orientation='landscape'` (or `portrait` if vertical requested), `resolution='1080P'`, `name='<scene> still'`, `entity_type='storyboard_images'`.
5. **Generate voiceover** with `generate_tts`:
   - No reference voice → `model='tts.design'` with `voice_instruction` describing a natural, conversational host voice.
   - Reference voice provided → `model='tts.clone'` with `reference_audio_url=<asset>`.
   - `name='<host> voiceover'`, `entity_type='voiceovers'`.
6. **Generate background music** with `generate_music` (`model='music'`, `name='<vlog> music'`, `entity_type='music'`) — subtle, non-intrusive, matching the mood; keep it low in the mix so dialogue stays clear.
7. **Generate video clips** with `generate_video`:
   - Host talking-to-camera / consistent identity → `model='ref2v'` with `input_urls=[<host still(s)>]` (reference mode preserves the person).
   - Single scene image → motion → `model='i2v'` with `input_urls=[<scene still>]`.
   - Audio-driven (uploaded audio as timing source) → `model='ref2v'` with `input_urls=[<host still>]` and `audio_url=<audio asset>`.
   - Pure text scene with no image → `mod