Stick Figure Studio · সর্বজনীন Open Skill
Stick Figure Studio
You can leverage this Skill to make a viral stick-figure video. It supports blending multiple art styles and enables your stick-figure character to perform actions as you specify across diverse scenes. You are welcome to post the completed video to your social channels.
ShortDrama
136 7 22
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-09-18/zxtzs5hb.jpg
--- name: stick_figure_studio display_name: Stick Figure Studio description: Use when the user wants to make a stick-figure / stickman video (one clip by default), or only generate stick-figure character images, or edit a single shot allowed-tools: - list_models - search_web - generate_image - recognize_image - upscale_image - generate_video - mix_clips - generate_tts - generate_music - fetch_video_info canvas: model_control: true --- # Stick Figure Studio No project database. Track progress with the chat, canvas assets, and `write_todos`. Keep an **internal** brief (title, type, ratio, Stick IDs, action, continuity, assets). For a video ask, also write an **internal content script** — that script is the only story/action source for `generate_video`. **Do not show the brief or the script to the user.** Character image prompts reuse the live `image_edit` skeleton (Nano Banana / Krea sentence order, t2i vs i2i, Keep vs change, self-check), then keep **stick anatomy** unless the user asked that a character not be stick-figure. **Character art style is unrestricted** for stick figures **and** non-stick characters — no default 2D, Rico, black-and-white, photoreal, 3D, or original-upload finish. If they named a look, **use theirs**. If they did not, do not force a finish. Color, gradients, shading, photoreal, and 3D on characters are all allowed. **Background / scene style is unrestricted** (any style). `style-bible.md` is a glossary only — do **not** apply a table row as the look. ## Understand intent (hard gate) Do **not** call `generate_image` or `generate_video` until the user's intent is fully understood — meaning, not keywords. Read what they want done, what must stay, and what success looks like. 1. **Parse the ask, not the nouns.** Infer video type, joke/message, who is on screen, mood, and whether they want a full film, a character sheet, or one edit. A short message still has intent. Do **not** infer a `style-bible.md` row from topic (fight → Pivot, explainer → whiteboard, skit → Rico). If they already wrote a **rough plot** (what happens, beats, a short script), **use it as written**. Do not rewrite, polish, reorder, add extra story, or "improve" it. Expand into a concrete plan **only** when they did not give a plot. 2. **Honor specified details.** If they named a look, accessory, color, shot, or reference, that is locked **in their words** (on the character unless they were talking about the set). Fill only the gaps they left vague (not the plot). Do **not** fill a missing look by applying a `style-bible.md` row. If they asked that a character is **not** stick-figure, do **not** convert them to stick anatomy. Their **art style is unrestricted** unless they named a look — do not lock them to the upload's original rendering style. 3. **Ratio: never ask.** Default **portrait `9:16`**. Use landscape `16:9` (or another Size they named) only if they already asked. Do not pause for this. 4. **Uploads and last assets are part of intent.** If they sent an image/video, `recognize_image` / `fetch_video_info` before planning. Do not ignore a reference they attached. 5. **Pin the target.** If more than one character, shot, or region could match, pick the obvious one from context. Do **not** stop to ask in the middle of making. 6. **Write a short internal plan** (never show it to the user, never wait for approval): Who / What happens / What to keep / Character style (**their named look**, or unrestricted if they did not name one) / Output. For a video ask, also write the **content script** in `01_plan.md` (beats, actions, dialogue, motion magnitude, tempo, opening/end state). Do not put a bible row in the plan. Then make it: character images (scene matched to intent; background any style) → **one** `ref2v` filled **from that script**. Do **not** generate keyframes unless they asked. 7. **Write motion magnitude (hard).** If the intent contains action, do **not** leave it as a bare verb, and do **not** rewrite it into a prep → o