Pixel Arcade Short · Open Skill 公开技能
Pixel Arcade Short
**Pixel Arcade Short · One Sprite, One Path-Stage, One 10s Playthrough · ref2v Native · Foley Kept** *The agent skill for creators who want a finished **8bit NES-style arcade playthrough**—not a photoreal live-action, a claymation, a LEGO brick film, a still portrait, a licensed-mascot lookalike, or a first-frame freeze of a sprite plate* **Pixel Arcade Short** is a production-native Agent Skill for Digen. It orchestrates a complete **reference-to-video (ref2v)** pipeline for one arcade episode: **Idea (or your real photo) → original sprite lock → empty path-stage still → clocked dope sheet → one 10s ref2v climb/run → pause on that clip** No storyboard grids. No contact sheets. No extra scene stills to “extend” the world. No TTS, no `audio_url`, no generated BGM. The lead stays an **original pixel sprite**. The stage still stays **empty of people**. The world **extends right or up inside the clip**. Speech is **off**—clip foley stays on if you ever join. The sprite plate is identity, not an opening keyframe. Play starts **already inside** the arcade screen. ## What you get | For creators | For the production stack | | --- | --- | | **Beginner-first** — no photo? the agent invents an original sprite from the idea you clicked and runs toward STAGE CLEAR | **Opinionated pipeline** — language lock, original-character gate, empty path-stage, duration=10, ref2v packing, follow-cam UP/RIGHT encoded in-skill | | **Photo-first** — your person / place / object is the lock; no photo is a generate path, not a dead end | **ref2v-first** — no accidental i2v / t2v / fl2v / director / first-frame chains; no physical model names | | **World that reads on sight** — chunk pixels, HUD chrome, four platform tops, flag on the last top, one through-path | **Two-ref binding** — `input_urls` = original sprite, then empty path-stage; neither still is the first or last frame | | **Dynamic ideas** — original hero + tileset + win-condition buttons written for this turn, never a red-cap plumber trio | **Six-section prompts** — `[reference generation]` first, then a compiled English clip body from the dope sheet, 1:1 clock rows | | **Delivery that continues** — after the 10s clip, buttons to redo / change sprite / change stage; longer cuts only if you ask | **Progressive disclosure** — craft lives in `references/`, not bloated context | | **Plain creator voice** — short lines about this sprite and this path; no pipeline jargon unless you ask | **Marketplace-ready** — clear triggers, defaults, and tool contracts | ## Pipeline | Stage | Output | | --- | --- | | **Intake** | Lock language + **9:16** + default **10s** (1×10s). Person photo → bind that face, restyle away from licensed mascots. Place photo → pixel the empty path. No photo + idea chosen → **one `t2i`** original sprite (look + 3/4 facing right, plain bg). Vague ask → clickable original-hero stages, then generate | | **Identity** | Sprite still as a **plain-background plate** (not standing in the stage). Upload path: `i2i` pixel conversion onto the person photo only. **No** `recognize_image` on generated stills | | **Set** | One empty **path-stage** (`t2i`, or `i2i` from a place photo, `locations`) — four tops P0→P3, HUD chrome, flag on P3, **no lead in the plate** | | **Dope sheet** | Clocked 10s sheet: axis **UP** (default) or **RIGHT**, hops **P0→P1→P2→P3**, coins, enemy, STAGE CLEAR. Identity, NES 3-pose hops, screenshot hero pose. No grids. No extra stage images | | **ref2v** | **One** 10s clip. Tool **must** pass `duration=10`. Prompt is the six-section pack; `summary` starts with `[reference generation]`. Sprite + empty path as **references**, not keyframes. Follow-cam along the locked axis; HUD pinned to the frame | | **Join** | **Skip** on the default 10s path. `mix_clips` only if you asked for 20s / 30s. Then `volume=1`. **No mute. No music tool. No BGM overlay.** | ## Production stack | Parameter | Default | Notes | | --- | --- | --- | | **Episode** | **10 seconds** | 1 clip × 10s. Optional 20s (2 clips) or 30s (3 clips) **only** if you ask | | **Engine** | `ref2v` · `reference_to_video` | Bare default channel. Forbidden: `i2v` / `ai2v` / `t2v` / `fl2v` / `director` / physical names such as `rm3.5-ref` | | **Clip duration** | **10 seconds** | **Always set `duration=10`.** Prompt text “10-second” does not set length. Omit → platform 5s | | **Video orientation** | Portrait · **9:16** | Landscape only if you ask. Sprite still, path still, and video share the lock | | **Image refs** | Original sprite always; empty path-stage always | Sprite = `t2i` `characters` (no person photo) or `i2i` from upload. Path = `t2i` / `i2i` `locations`. **Exactly two stills** on the default path | | **Ref order** | Sprite → empty path | Identity plate, not a first-frame. Forbidden: empty `input_urls`; single-image calls; empty path as opening frame; frozen replica of either still; a third scene still to “extend” the world | | **Identity plate** | **On, every no-upload / pixel-convert still** | Original silhouette, 3/4 facing **right**, **plain studio background**. Forbidden: sprite already inside the stage; Mario / Luigi / any licensed mascot | | **Path lock** | **One empty 4-top path** | P0 spawn → P1 → P2 → P3 (flag). Default axis **UP** (portrait height). **RIGHT** only for a ground run. Forbidden: a single floor with one gap; a hero painted into Picture 2 | | **World extend** | **Inside the clip** | Follow-cam along the locked axis. Same tileset continues; extra same-tileset ledges may enter. HUD pinned to the **frame** top. Forbidden: locked one-jump crop; extra stage images | | **Lead lock** | **Same original identity after lock** | Face mass, palette, one unique accessory stay. Forbidden: swapping people, copying a commercial sprite, aging a child, sexualizing anyone | | **Asset QA** | **Off** | Do **not** `recognize_image` generated sprite or stage stills. Wait for URIs and continue | | **Speaker lock** | **N/A — no speech** | No dialogue, no VO, no `generate_tts`, no burned-in captions | | **Dialogue** | **Off** | Story is visible verbs (walk, hop, collect, stomp, touch the pole), not subtitle jokes | | **Picture audio** | **Clip foley, kept** | Light 8bit SFX from the video. If you join, the join does **not** mute. No `generate_music` | | **Cut grammar** | **1×10s, no join** | Longer cuts: `mix_clips` `volume=1`. Forbidden: mute join; second pass overlay BGM | | **BGM** | **Off** | No music tool. Do not ask the video model to invent a score | | **Mini-script** | **Clocked dope sheet** | ≥18 rows: clock \| boots on P# \| NES pose \| prop/HUD. Failed board = one gap then stand on the same floor | | **Confirm** | **After the 10s clip** | Default pause so you can redo / change sprite / change stage. One-shot / generate-all still pauses on that clip | | **Buttons** | **Written this turn** | Not a fixed catalog. Ideas, stages, and next moves named from this original sprite and this path | | **Language** | **User-typed lock** | Replies, buttons, todos, asset names in your language; image/video prompts stay English | ## Modes | Mode | You bring | Agent delivers | | --- | --- | --- | | **Full auto-run** | “Make a pixel arcade short” / generate all | Invent an original sprite + empty 4-top path if needed → stills → dope sheet → one 10s follow-cam clip | | **Idea-first** | One-line premise (“forest climb, fox-ear hero, flag at the top”) | One `t2i` original sprite → empty path → clocked sheet → clip | | **Photo-first** | Your person / place / object photo | Person: bind that face, `i2i` to an original pixel sprite. Place: pixel the empty path, strip every person. Object: pickup in the **video**, not a person in the stage | | **Pick-an-idea** | Nothing locked yet | 2–4 concrete original-hero + path + STAGE CLEAR buttons in your language; click → generate that sprite | | **Set swap** | “Make it a neon rooftop run instead” | New empty path-stage; same locked original sprite | | **Axis swap** | “Go right, not up” | Lock **RIGHT**; same sprite; new left-to-right four-top still; follow-cam right | | **Single-clip** | Default — “Just 10 seconds” | One ref2v clip with `duration=10`; skip the join | | **Single-stage edit** | “Redo the clip” / “Change the forest” / “New sprite” | That stage only, then remake the affected clip | | **One-shot** | “Finish in one go / no confirm” | No pause between stills and video; still pause on the finished 10s clip | | **Lengthen** | “20 seconds” / “30 seconds” | 2 or 3 × 10s chapters, extra empty stages only then, then one unmute join | ## Example prompts **Full auto-run** > Make a pixel arcade short. Portrait, 10 seconds. **Idea-first** > Forest climb. An original fox-ear sprite hops up four dirt platforms to a flag. 9:16. **Photo-first** > This is me. Turn it into an original 8bit hero climbing a dungeon stair well. Portrait. **Pick-an-idea** > I don’t have a photo. Give me a few arcade-stage ideas and just start. **Set swap** > Same sprite. Change the path to a neon rooftop run, camera going right. **One-shot** > Generate everything in one go. Don’t stop after the stills. **Single-stage edit** > Redo the clip — it should hop platform to platform upward, not one jump on a locked floor. ## When to use - Portrait 8bit / pixel / retro arcade / NES-style playthroughs: one original sprite, one empty path, one 10s climb or run - You uploaded a real photo of a person, place, or object, **or** you only typed a scene - You want to change the sprite, rewrite the dope sheet, redo the clip, or (only if you asked) remake a longer cut - Creators who want a finished 10-second arcade screen with original clip sound — follow-cam up or right, not a mute beauty reel and not a licensed-mascot screenshot > **Not the right tool when** you need photoreal live-action, claymation, LEGO brick films, 2D anime, a product ad that must stay photoreal, a spoken / lip-sync track, a generated song overlay, storyboard grids for client sign-off, a Mario / Luigi / other licensed sprite, people painted into the stage still, or real-world harm. ## Invoke ``` /pixel-arcade-short ``` Or: *"Turn this into a 10-second portrait 8bit arcade playthrough — original sprite if I uploaded a face, one empty four-platform path, hop up or right with follow-cam, ref2v, duration 10, keep the clip sound, no mute mix, no extra scene stills."* **Your photo · Your path · One original arcade world.**
ShortDrama
23 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3325120/2026-09-20/h4d7d3dn.png
--- name: pixel_arcade_short display_name: Pixel Arcade Short description: Use when the user wants an 8bit pixel retro arcade playthrough (original sprite hero, NES-style side-scrolling stage, HUD, jump-and-collect, stage clear), with or without an uploaded photo; also use to change the sprite, stage, or redo the clip. allowed-tools: - recognize_image - generate_image - generate_video - mix_clips - create_document - update_document --- # Pixel Arcade Short Make a **portrait (9:16) 8bit arcade clip**. Default output is **one original sprite + one empty path-stage still + one 10-second video**. The still is the **start of a path**. The clip **extends that same tilemap right or up** (follow-cam) so the player hops **platform to platform**, not one jump on a locked floor. This Skill has **no** structured project state. Keep facts in the reply or `create_document`. Reuse canvas URLs. **Video channel (hard):** every clip uses `generate_video(model='ref2v', gen_mode='reference_to_video', duration=10, orientation='portrait')` — the **bare `ref2v` default channel**. **`duration=10` is a required tool argument.** Writing "10-second" only in the prompt does **not** set length. If `duration` is omitted, the platform default is **5 seconds**. Forbidden: omitting `duration`; passing `duration=5`; `i2v` / `t2v` / `ai2v` / `fl2v` / `director` / physical model names such as `rm3.5-ref`. `gen_mode='reference_to_video'` must be passed with `ref2v`. `input_urls` = **original hero sprite, then the empty stage still** (exactly 2 images). Never put an audio URL in `input_urls`. This is **reference-to-video**. The prompt is the **six-section pack** in `references/04-video.md`; `summary` **must start with** `[reference generation]`. Forbidden: a 5-sentence mood blurb; treating any input still as the first/last frame; `[image to video]`; first-frame / keyframe completion. **World representation (hard):** | Layer | What it is | What it is not | |---|---|---| | `<Picture 1>` | **Original** identity sprite, 3/4 side facing right, plain background | Mario / Luigi / any licensed mascot; a game frame | | `<Picture 2>` | **Empty path-stage**: same tileset, **4 unlabeled walkable tops** going **up** (default, 9:16) or **right**, HUD chrome in the top band, flag pole on the last top. **No person, no platform labels** | A screenshot with a player; a single floor with one gap; `P0`/`P1` painted on steps | | Video prompt | 10s **path adventure**: follow-cam, **zigzag** hops (UP-RIGHT / UP-LEFT), at least one **backstep + charge**, coins, enemy, STAGE CLEAR | Locked camera, one vertical hop, three extra scene stills | `ref2v` locks appearance. It does **not** collide. The player exists only in Picture 1 and is **placed by the video prompt**. Picture 2 must stay empty of characters so the model does not copy a second hero (or a licensed one) out of the stage. **Original character (hard):** do not copy a commercial sprite. Forbidden looks: Mario / Luigi (red or green cap + mustache + blue overalls), Sonic, Link, Pikachu, Pac-Man, Mega Man, Kirby, or any named licensed mascot. Invent a unique silhouette + palette + one accessory that is **not** that combo. Put that in the **image prompt**. Do **not** run `recognize_image` on generated stills to check or reject them. **Default length (hard):** **10 seconds = 1 clip**. Generate **one** stage still, then **one** `generate_video`. Do **not** generate 2–3 stage images or 2–3 clips unless the user **explicitly** asks for 20s / 30s / another room. **Zero CJK in this package.** Skill files stay English (handbook only). User-facing runtime copy follows `user_lang`. Image and video `prompt` fields stay English. Image prompts follow packing in `references/02-assets.md`. Video prompts follow the long six-section pack in `references/04-video.md`. ## Global Language Lock (mandatory, highest priority) Follow the **user's typed sentences**, not platform chrome. **Chrome trap:** strip the slash-invoke line, `<untrusted_