Ceiling Hero Shorts · सार्वजनिक Open Skill

Ceiling Hero Shorts

**Ceiling Hero Shorts · Theme to Look to Hang-Up Cut · rm3.5-ref Native · Preset-Audio Driven** *The agent skill for hosts and creators who want a finished **ceiling talking selfie Short**—not a cinematic crawl montage, a mute superhero trailer, or speech invented inside the video prompt with no real audio track* **Ceiling Hero Shorts** is a production-native Agent Skill for Digen. It orchestrates a complete **reference-to-video (`rm3.5-ref`)** pipeline for one viral ceiling-hang talking clip: **Idea → performance intent → indoor ceiling setup → one identity still → compact English hang-up prompt → 10s audio-driven ref cut → mixed Final Cut** No storyboard grids. No contact sheets. No per-shot i2v chain. No six-clip stitch. No TTS. Speech lives in the **Skill preset track** (`global_fixed_audio`), passed as `audio_url`. The video model lip-syncs to that track. One clip, one video—standing selfie, one jump, one-hand full-palm ceiling hang inside a single 10s render. ## What you get | For creators | For the production stack | | --- | --- | | **Beginner-first** — no script? the agent picks emotion, room, and creator energy and runs to the cut without stopping | **Opinionated pipeline** — locked 10s hang-up flow, one look plate, one preset dialogue track, compact prompt encoded in-skill | | **Talking-host first** — the hook is a real person facing the camera, then the ceiling reveal, not a silent action mood board | **ref-first** — `rm3.5-ref` + `reference_to_video`; no accidental i2v / ai2v / t2v / storyboard / per-shot video chains | | **Identity of face and suit** — one chest-up character plate locks the host across the whole clip | **Audio-driven cut** — preset `audio_url` only; never `generate_tts`; never put audio inside `input_urls` | | **Living hang-up grammar** — stand and talk → one jump → one palm flat on the ceiling → keep talking with the free hand | **Compact prompts** — front-load + four time windows + anatomy / hang lock / free hand, each constraint once | | **Delivery that continues** — after the cut, buttons are **Regenerate video / Change emotion or room / Done**, already in your language | **Progressive disclosure** — craft lives in `references/`, not bloated context | | **Plain host voice** — short lines about the vibe, the room, and the clip; no pipeline jargon unless you ask | **Marketplace-ready** — clear triggers, defaults, and tool contracts | ## Pipeline | Stage | Output | | --- | --- | | **Intake** | Lock language + **9:16** + **10s** + **single host** + one hang-up episode (no orientation question) | | **Performance** | Emotion, talking energy, free-hand gestures vs the preset vocals — not a TTS script | | **Ceiling setup** | One indoor room (bedroom default); furniture gravity normal; body contradicts gravity after the jump | | **Look** | One vertical character plate (`t2i` / `i2i`). User face or cosplay upload beats AI when provided | | **Prompt** | Compact English hang-up prompt: timeline + anatomy + locked ceiling palm + free hand at chest | | **ref2v** | One look ref + preset `audio_url` → **10s** `Talking Clip`. Lips follow the preset track | | **Mix** | Same preset audio mixed onto the clip → **`Final Cut`** | | **Sequel** | Next clip in this thread — reuse look when unchanged, new room/emotion, one new 10s video | ## Production stack | Parameter | Default | Notes | | --- | --- | --- | | **Episode** | **10 seconds** | One `rm3.5-ref` video. Film uses 0.00–10.00 of the preset track. User asks longer → more 10s chapters | | **Engine** | `rm3.5-ref` · `reference_to_video` · `generation_type='reference'` | Identity lock from one still. Forbidden: `i2v` / `ai2v` / `t2v` / `fl2v` / director | | **Clip duration** | **10 seconds** | Always set. Hang-up windows live inside the clip, not as separate videos | | **Video orientation** | Portrait · **9:16** · **480×864** | Landscape only if you ask. Look plate stays vertical even if the cut is landscape | | **Enhance skip** | **`skip_prompt_enhance=true`** | Mandatory. Keeps the hang-up timeline and palm lock from being rewritten | | **Presenter** | **Single host** | Real-person spider-pattern **body** suit; **bare head** unless you ask for a mask | | **Image refs** | One `Hero Character Ref` required | Neutral standing portrait, empty hands. No hang stills, no pose boards, no grids | | **Ref order** | Character look only in `input_urls` | Motion, room, jump, and ceiling hang come from the **video prompt** | | **Speech** | **Preset-audio locked** | Resolve `global_fixed_audio`. Pass the URL as `audio_url`, `audio_start_time=0`. Forbidden: `generate_tts`, `generate_music`, user-upload song, mute ref2v, audio inside `input_urls` | | **Timbre** | **Skill preset vocals** | Do not design or clone a voice. Do not ask the user to upload music | | **Framing lock** | **Chest-up MCU** | Unseen phone selfie. Face large in frame. Forbidden: knees, wide, back view, walking around the room | | **Hang-up flow** | **Locked four windows** | 0–1.50 stand and talk · 1.50–3.50 one jump, right palm locks on ceiling · 3.50–8.50 hang, palm stays locked, left hand small gestures · 8.50–10 hold | | **Ceiling contact** | **One full palm, then locked** | Palm faces the ceiling, not the camera. After lock: that hand does not move. Exactly two hands | | **Dialogue** | **On, from the track** | Lip-sync to `<Audio 1>`. No on-screen subtitles / captions burned into pixels | | **Cut grammar** | **Inside one clip** | Time windows in one 10s render. No storyboard grids, no per-shot i2v chain, no default stitch of action beats | | **Confirm** | **After delivery** | First clip auto-runs; pause only if you ask, then buttons after the video. Never pause to approve the still | | **Series** | **Next clip** | Reuse look when unchanged; new performance + prompt + cut each time | | **Language** | **User-typed lock** | Replies in your language; image / video prompts are English; no trademark names in model prompts | ## Modes | Mode | You bring | Agent delivers | | --- | --- | --- | | **Full auto-run** | Nothing — or “make a ceiling talking selfie” | Invent vibe + room → look → hang-up cut on the preset track | | **Idea-first** | Topic + one-line premise | Expand to performance notes → full pipeline | | **Face-driven** | A person photo + brief | Bind the upload as the look; do not redraw the face you uploaded | | **Cosplay-driven** | A suit / identity photo + brief | Lock wardrobe from the upload; hang-up motion still comes from video | | **Room-driven** | A bedroom / apartment photo + brief | Keep furniture gravity normal; host still hangs after the jump | | **Single-stage edit** | “Change the room” / “Redraw the host” / “Regen video” | That stage only, then remake the cut | | **Fix the hang** | “The ceiling hand moved / deformed” | **Regenerate video** with the same compact prompt (new seed). Do not stack extra prompt text | | **Same-thread sequel** | “Next clip” / “same host, new vibe” | Continue from end-state · look reuse · one new 10s video | | **Landscape on request** | “16:9 / horizontal” | Override the 9:16 default for this run only | | **Mask on request** | “Put the mask on” | Override the bare-head default for this run only | ## Example prompts **Full auto-run** > Make a portrait ceiling talking selfie, 10 seconds, spider-suit host, face to camera. **Idea-first** > Chill hang after a long day. Bedroom. Playful energy. **Face-driven** > This image is the host. Build a 9:16 ceiling hang talking Short from it. **Cosplay-driven** > Use this suit photo. Same person, ceiling selfie, one hand on the ceiling after the jump. **Room-driven** > This is the bedroom. Keep the furniture right-side-up. The host hangs after one jump. **Fix the hang** > The palm looked right at first, then the ceiling hand moved. Regen video, same prompt. **Series** > Same host, new room — living room, smug energy, another 10-second clip. **Landscape override** > Same host, landscape 16:9, 10 seconds. **Mask override** > Same flow, but the host wears the mask pulled up so the face still shows. ## When to use - Vertical talking-head **ceiling hang** Shorts: stand, jump, one palm on the ceiling, keep talking - One-line vibes, a face / cosplay / room photo, or “just make the clip” - Creators who want a **real preset vocal track** driving lip-sync, not prompt-only speech and not TTS - Spider-pattern **body-suit** hosts with a **bare face** to camera (mask only if asked) - Anyone tired of cinematic crawl montages, mute superhero trailers, or four hard-cut action beats stitched after the fact > **Not the right tool when** you need a 60-second six-clip mix, a product-only pack-shot ad with no on-camera host, a generated TTS voice, lip-sync to a user-upload song, first-person wall-crawl POV, storyboard grids for client sign-off, horror / gore, prompt-only in-clip speech with no `audio_url`, both-hands spider-crawl on the ceiling as the whole clip, or a drama series with many unrelated locations per minute. ## Invoke ``` /ceiling-hero-shorts ``` Or: *"Turn this into a 10-second ceiling talking selfie — face on camera, one jump, one palm locked on the ceiling, preset vocals, one ref cut."* **Your host · Your room · One coherent ten seconds on the ceiling.**

@jimi

ShortDrama

49 3 28

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3325120/2026-08-21/4nz2uaz9.jpg

---
name: ceiling-hero-shorts
display_name: Ceiling Hero Shorts
description: Use when the user wants a Spider-Man / Miles Morales ceiling talking selfie Short (real-person cosplay, bare face to camera, stand-and-talk then jump into single-hand full-palm ceiling hang with load-bearing contact, lip-sync to preset vocals, chest-up phone selfie framing, TikTok/Shorts creator style). Also use when advancing or editing one pipeline stage only.
allowed-tools:
- generate_image
- generate_video
- mix_clips
- list_models
---

# Ceiling Hero Shorts — Spider-Man Ceiling Talking Video

**One content type only:** a real-person Spider-Man / Miles Morales **style creator** films a phone selfie talking Short: **stands chest-up talking → jumps → one hand full palm flat on ceiling directly above head (all finger pads + fingertips, load-bearing hang) → keeps talking with the free hand**. Lip-syncs to the Skill preset track vocals. **Not** a cinematic action scene, crawl, flip, or fight.

**Dialogue lives in the preset music** (`global_fixed_audio`). **Never** generate separate TTS.

No structured project state — progress is conversation + canvas assets + `write_todos`.

## Global Language Lock (Mandatory, Highest Priority)

[Language Protocol] Follow the language of the user's input; use that language throughout for replies, performance plans, progress notes, error messages, delivery summaries, review conclusions, and **every `set_guidance` button label**.

### 0. Initial Language Lock (Init; before first reply)

- On the user's **first message**, immediately lock `user_lang` and use it from the **first word** of the first reply onward. Greeting in one language then switching is **forbidden**.
- This Skill document and **all reference branch files** are **English only** — **no Chinese characters anywhere in skill documentation** (frontmatter, rules, examples, negatives, templates). Runtime user-facing text uses **`user_lang`**; do not embed non-English sample copy inside skill files.
- Before **every** user-facing output, self-check `user_lang`. If `user_lang` is English, include **no Chinese characters** in user-facing text. If `user_lang` is Chinese, user-facing text stays Chinese — do not inject English labels unless they are proper nouns the user used.
- Do **not** switch language because branch files are written in English.
- Do **not** switch language mid-run unless the user clearly switches first; then lock the new language for all subsequent user-facing output.

### Model Prompt Language (Mandatory; separate from user-facing)

| Layer | Scope | Language |
|---|---|---|
| **USER-FACING** | Replies, progress, plans shown to user, errors, delivery copy, **`set_guidance` button text** | **`user_lang` only** |
| **IMAGE PROMPT** | `generate_image` prompts | English structural keywords + project content |
| **VIDEO PROMPT** | `generate_video` prompt body | **English always** |

- Global Language Lock **does not** authorize a non-English video prompt when `user_lang` is Chinese (or any other language).
- **IP-safe wording:** user may say Spider-Man / Miles Morales in `user_lang`; model prompts use descriptive cosplay tokens — **no trademark names** in image/video prompts.

### Internal vs User-Facing (Mandatory)

- **INTERNAL ONLY** (never show): validation checklists, Asset ID tables, quality gate logs, tool/model/channel names, raw prompt packs, Beat-Sync internals, preset URI details.
- **USER-FACING:** short progress in `user_lang`; **`Talking Clip` shown right after video gen**; **`Final Cut` after mix**. No URL dumps. **Do not pause to review the character still.**

## Pause & Guidance Protocol (Mandatory)

**Rule:** whenever the workflow **pauses** and waits for the user, the same final turn **must** call `set_guidance` with **2–4 contextual buttons in `user_lang`**. **Never** end a paused turn with text alone.

### When to pause + guide (must `set_guidance`)

| Pause reason | Guidance examples (translate to `user_lang`) |
|---|---|
| **D