---
name: youtube
description: "Reference guide for a YouTube channel with a recurring AI character — a long-form 16:9 video where the character hosts to camera, or a solo video podcast. What to make from vidIQ research (breakout videos on the topic, trends, the person's channel), a reference episode's structure from its transcript, the person's own script, the host on MiniMax H3 Max Turbo (stub + extend chain, output continuation, two camera angles), title, description, chapters and thumbnail ready to paste. Classifies the job and hands off to vidiq research, avatar-podcast, avatar-long-speech and reels-voice. Use when the user says YouTube, channel, long video, podcast, video podcast, episode, host, talk show, thumbnail, title ideas, what's trending, ютуб, канал, подкаст, выпуск, ведущий, обложка, идеи для видео."
---

These creative references describe source experiments with external tools. Use your own supported account and tools; this download installs no automation. Settings and timings are experimental observations, not current FlashBoards availability. Use only media and voices you have permission to use.

# youtube · reference workflow

**Work for the person's own character: their look sheet / voice from their character card. The example run belongs to a demo character.**

Slots: {character} · {look sheet} · {voice} · {topic} · {studio} · {reference} (a breakout episode found with vidIQ).

## 🏆 Source experiment pick — an episode like a breakout on the topic

1. **What to make** — vidIQ on the person's account (using their supported tools): `outliers` on {topic}
   (breakouts: far more views than the channel usually gets), `keyword_research` for demand. Offer at most 3 ideas, one line each.
2. **Reference** — the chosen breakout's `video_transcript`: take the structure (welcome → hook → turn → images → one small thing
   to do → goodbye) and the length, never the text.
3. **Script** — the person's own words in that structure, in segments of ≈ 45 s of speech.
4. **Voice** — ElevenLabs v4 with {voice} (`reels-voice`), one file per segment.
5. **Host video** — `avatar-podcast`: two start frames of {studio} (medium + close-up, 16:9), one chain per segment on
   MiniMax H3 Max Turbo with **output `continuation`**, the angles alternating, glued with ffmpeg.
6. **Packaging** — `generate_titles` (vidIQ), a description with chapters from the segment start times (`0:00 …`), and a
   thumbnail drawn from {look sheet} (face big, ≤ 5 words); vidIQ `generate_thumbnail` only for the idea and the words.
7. **Publish** — the person uploads the file on YouTube; give the title, description, chapters and tags ready to paste.
   This download supplies no uploader; use your own supported YouTube upload workflow.

**Alternative quality route in the source: Seedance 2.5** reference-to-video — 30 s per call with the same prompt (`avatar-podcast`), at 720p. The measured comparison was a product commercial; compare it on your own material rather than treating that preference as a proven result for every format.
At 768P on H3, a minute of host footage took ≈ 4 min of generation; independent segments can run in parallel.

## Classify

| The job | Go to |
|---|---|
| Ideas, what's working on the topic, the person's channel numbers | vidIQ: `outliers`, `trending_videos`, `keyword_research`, `channel_analytics`  |
| A solo podcast or a host talking to camera, 16:9, a minute and longer | `avatar-podcast` |
| One segment, one angle, standing or seated | `avatar-long-speech` |
| A Short for the same channel (9:16, ≤ 15 s) | `reels` |
| The voice: pick, clone, emotion | `reels-voice`, `reels-voice-clone`, `reels-emotion-tags` |
| A thumbnail with the character | the image tools with {look sheet}; words from vidIQ |

## Global rules

1. **Structure from the reference, words of your own.** Never reuse a reference's sentences, title or thumbnail as is.
2. **Extend with `continuation`**, never `extended` — `extended` softens the picture step by step.
3. **Cut to the other angle every 40–50 s** of speech: a fresh stub per segment keeps the face sharp.
4. **Publish the clips' own sound**, never the original voice files under the video — the lips follow the clip.
5. Keep research bounded: one pass per question, ≤ 10 results, one thumbnail per ask.
6. Never reuse the example run's files, prompts, voices or images for a person.

## Example run (reference only — never reuse its files, prompts, voices or images for a person)

- Demo character Mara Vale, «Slow Talk» episode 1, 9.10.2026 — see `avatar-podcast` (reference: a breakout of 62× on a 16k channel,
  96 s in two segments).