---
name: yujeong-2x-shortform
description: 'Use when making a new "What if X doubled every time you ...?" absurd-escalation short (9:16 block format, voxel character, male narration, end card tailored to the episode) or fixing an existing episode. Example triggers: next episode of the doubling series, absurd escalation short, new Yujeong episode, fix the 2x episode, next doubling episode, fix the 2x episode pacing.'
---

# Yujeong doubling series short-form production

When to use: follow this only when making a new exaggeration-gag short where "X doubles each time" or fixing an episode of this series. For general ads or other formats, ignore it and follow youvico-production-pipeline. Answer and report in the user's language. UI names below are given in English; if the app shows them in another language (for example Korean), find the item with the same meaning.

## Ask the user before starting
The point of this skill is not to remake the example but to make the user's own video. Before starting, ask the questions below together, in one message.
- Skip anything the request already answers.
- If the user says "you decide" or "use the defaults", go with the defaults in parentheses.
- If the user seems to have no idea yet, briefly suggest 2–3 options that differ from the example.
- Once answered, show a one-line summary of what was decided, then start.

1. What doubles? One line: "What if ~ doubled every time you ~?" With no idea yet, the agent suggests 2–3.
2. Which lead character and art style? Point to the character reference image. The file IDs in this skill belong to the example project. (default: the project's character sheet, voxel style)
3. Who delivers the punchline? (default: someone who fits the topic (mom, team lead, vet and so on))
4. Narration voice and language? (default: male narration in the user's language)
5. Which brand and tagline go on the end card? (default: a tagline tailored to the episode; if the original end card can't be loaded, render a new one with the same layout)
6. Length and budget? (default: 30–45 s, show the estimate first)

## 1. Standards (EP1 "feedback doubles" = the benchmark episode)
- Length 30–45 s = main part 25–40 s + end card 4.5 s. An end card over 10 s is a failure.
- Structure: hook (0–4 s, a question like "What if ~ doubled every time you ~?") → round 1 (still fine) → 4–6 steps of rapid escalation (the real-world comparison keeps getting bigger) → twist punchline (a short line from mom / team lead / vet etc.) → end card.
- One scene change every 2.5–4 s. If a shot looks frozen for more than 5 s, it fails. Zero still-image shots (all video).
- Line–picture match: the thing the narration names must be on screen at that moment. Scene change complete = start of that line +0–0.2 s. A next scene that appears before the line also fails.
- Fill the frame (required): a 720p video clip left as is on a 1080×1920 canvas gets a black border around it. Set every video clip to set_layer_transform scale 1.5. The end card (rendered at 1080×1920) stays at scale 1.
- The brand (YOUVICO) goes only on the end card. No brand mention in the main part's lines or picture.
- End card copy is tailored to the episode: a strikethrough chip "exaggerated final value → real value ✓" + a two-line tagline "Not 〈final exaggeration〉, / finish it with 〈solution〉."
- Block format (1080×1920): hook 80px white, centered, y0.19 (14 characters or fewer per line) / label "Nth ○○" 48px y0.305 / counter 190px y0.375 / badge 44px white text on #E5484D background y0.447 / subtitle 66px #FFD43B y0.76. Left-aligned elements x=(63+width/2)/1080 (width approximation: counter digit 110 · Korean character 190 · space 50px, label digit 29 · Korean character 48px, badge Korean character 44px + 56px side padding; for Latin letters, about 0.6× the font size per letter). The top 30% is the hook's place, so specify "top 30% plain wall" in keyframe prompts.
- Sound: video clips 60% (changing volume on a clip with no audio fails → ignore), BGM 22%, pop SFX 45% (when the counter changes), logo SFX 60% (start of the end card).

## 2. Steps
1. Script: actually calculate each round's number and make a table (2^n, unit conversions, measured values of the comparison objects). The punchline is a line that brings it back to the first target's original value. End with the end card tagline sentence.
2. Narration: ElevenLabs "Yong Gyu" (Eleven v4). Enter the whole script as one line (with line breaks it generates only the first paragraph). After entering it, reread for typos. external_page show → Download → wait_for_downloads → upload_project_attachment.
3. Timing: get segment times with run_local_ai transcribe(attachmentId, saveToProject=false) and make a beat table. If numbers are transcribed wrong, correct them from the script and calculation table.
4. Keyframes: Nano Banana Pro (gemini-3-pro-image), 9:16, first image + character reference FXUXCNL73FWD9PHI. Each next image takes the previous one as the first reference: "Keep EXACTLY the same … change only …". "No text, no letters, no logos". Submit an episode's 7 images at once (14 at once for two episodes).
5. Video: Kling 3.0, 720p, 9:16, generateAudio=false. Consecutive scenes use a [Ki, Ki+1] first/last-frame chain (5 s; the hook shot 8 s); the last reaction shot is a single image, 5 s. Submit them all at once.
6. Measure transition points (required): render all clips as one frame sheet (row = clip, column = source 0.3–4.9 s) and note for each clip the "source second when the next scene is complete". Kling chains usually finish the transition at source 1.5–3.4 s.
7. Edit (the fast way): "Duplicate this edit" on the previous episode's edit (a duplicate keeps the same clip IDs) → delete video, narration and end card → change only the text content with update_layer_text and reposition to the beats with trim_clip (start/end) (from the last scene backward). Recalculate the left-aligned x with set_layer_transform.
8. Video placement (recommended method): append to V1 in order with add_clip_to_timeline, then put set_clip_speed (0.75–1.6) on each clip so that "transition complete = next line start + 0.1 s". If needed, cut the front of the source with trim_clip. move_clips locks after 5 overlap errors, so don't use it to move clips. Then set all video clips to scale 1.5.
9. Audio · end card: narration on A3, logo SFX at the end card start, BGM cut at the end card's end. The end card is render_motion_graphics (1080×1920, 4.5 s) — load the original renderId b4095171-8c2a-4e75-8029-5a1038d07300 (app icon FXWNMILXT8AUWZ2G) with get_motion_graphics_source, reuse the sourceId and replace only the copy. The end card starts right after the last line ends (punchline end +0.2 s). The back part of the narration (the tagline) overlapping the end card is intended.
10. Export: export_draft to the finished folder. The new version is review; previous and failed versions get the reason at the end of their name (e.g. _screen-border) and closed.

## 3. Checks
- Frame sheet of the result video: keep the 9:16 ratio, 10–14 points — 0.5 s, each line start +0.5 s, the punchline, the middle of the end card. Use a loud background color like magenta (#f0f) and draw with object-fit:fill so black borders show.
- Pass: no border, each point's picture matches what that moment's line names, hook and counter not cut off, end card 4.5 s with copy matching the episode, brand only on the end card.
- Render 1–2 suspicious scenes large at 540×960 and read the text.

## 4. When it fails
- Black border → change V1 video clips to scale 1.5 and re-export as v(n+1). Costs no credits.
- Next scene appears before the line → realign the transition points with steps 6–8.
- When reusing a finished file with baked-in text, punch in 1.05× or less.
- If update_layer_text doesn't work through apply_changes, use individual update_layer_text calls.
- If stuck on "Waiting for the service lookup response", make the failed lookup succeed again with the same requestId. For 429, rest a few seconds and go one at a time.

## 5. Report format
Per episode: file name · ID · length / beat table / points checked / what still needs a person to check / credits (separate estimate and actual charge, sum of billedCredits from get_ai_generation_status). Actual charges for reference: image 15, Kling 720p 5 s 32 · 8 s 51 → one episode (7 images + one 8 s + six 5 s) ≈ 348. Estimates come out higher than this.

## 6. Examples (absurd topics)
- Ramen water doubles: 1.1 L → 4.4 L pot → 282 L bathtub → 2,307-ton swimming pool → 9.45-million-ton Seokchon Lake → 1.3 billion km³ of Earth's oceans → mom: "The noodles are all soggy." End card "Let the noodles get soggy, / but finish work on time."
- 5-minute alarm doubles: 10 min → 2 h 40 min → two days → a month → 1 year 3 months → 319 years → team lead: "You're 5 minutes late." End card "Not five more minutes, / finish it now."
- Cat treats double (4 kg×2ⁿ): 8 kg → 128 kg sofa collapses → 4-ton elephant → 131-ton blue whale → 130,000-ton aircraft carrier → a line from the vet. End card "Not 25 times, / finish it in one go."
- Laundry mountain doubles (30 cm×2ⁿ): 60 cm → 2.4 m chair height → 19 m apartment → 307 m Han River bank → above 9.8 km Everest → a single-sock punchline. End card "Not Everest, / finish it today."

Change log: v1.0 2026-10-28 first written / v1.1 added transition-point measurement, edit duplication and x width approximation from EP8·9 production / v1.2 2026-10-28 added from EP10·11 production and the EP8·9 border fix: scale 1.5 required, matching timing with speed (no move_clips), border check method, end card original renderId, one-episode cost / 2026-10-30 added questions before starting