Animal Adaptation Shorts · publiczny skill Open Skill
Animal Adaptation Shorts
Animal Adaptation Shorts to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
Anime
3 1 6
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3761915/2026-10-19/bb35du7s.jpg
# Animal Adaptation Shorts Produces one finished 9:16 short-form nature documentary (~10 seconds) about a single remarkable animal adaptation: research → beat sheet → keyframes → animated beats → voiceover → score → final mix with burned-in text overlays. ## When to use - The user asks for a "mind-blowing animal adaptation" clip, wildlife / nature reel, or short-form documentary about an animal trait (speed, strength, senses, camouflage, mimicry…). - The user supplies a prompt in the shape of a beat sheet (timings, text overlay lines, mood words like "deep narrator", "dramatic low bass") — treat it as the spec and fill the gaps. - The user asks to continue or redo one of these videos in a follow-up turn. Do not use for: general talking-head clips, non-animal topics, or requests to edit a video the user already uploaded (that is a different workflow). ## Reference files - `references/beat-sheet.md` — the beat/overlay template, a fully worked example (mantis shrimp), and the rules for writing overlay + narration lines. **Read this before generating anything.** - `references/style-library.md` — copy-ready prompt language for keyframes, motion, voice and score, plus per-habitat variations (ocean / insect / bird / mammal / deep sea). - `references/tiktok-captions.md` — the optional word-by-word caption pass (`caption_video`): templates, parameters and when to offer it. Read it only when the user asks for TikTok-style captions or when you are about to offer that option. ## Steps Use `write_todos` to list the stages, then update it as you progress. ### 1. Lock the subject and the beats Extract from the user's request: the animal, the adaptation being shown, the total duration (default 10s), the beat timings (default 0–3s / 3–7s / 7–10s), the overlay lines, and the mood words. If the user gave timings or overlay text, keep them **verbatim** — do not rewrite their lines. If the request is vague (just an animal name), pick the single most astonishing documented adaptation and design the beats yourself. ### 2. Research the fact (2 quick searches) Call `search_web` (1–2 queries, e.g. "<animal> adaptation how it works speed strength") to confirm the claim behind the overlay lines. Only put in the overlay + narration what the search supports. If a search contradicts the user's line, keep the user's line but pick a supportable number, and mention the discrepancy in your final message. Never invent precise figures (velocities, force multiples, wavelengths) that no source supports. ### 3. Build the beat sheet Read `references/beat-sheet.md` and write the three beats with: purpose, camera framing, motion, overlay text (or "none"), and the narration line. Keep the total narration to ~2 short sentences that fit the 10s window. ### 4. Generate one keyframe image per beat `generate_image` with: - `model='t2i.hd'`, `orientation='portrait'`, `resolution='1080P'` - `input_urls=[]` (text-to-image; add a reference image only if the user supplied one) - `name` = "<animal> — beat <n>: <short label>" e.g. "Mantis Shrimp — beat 1: eye macro" - `entity_type='storyboard_images'` - Prompt = the camera/framing line + style line from `references/style-library.md`, plus "vertical 9:16, cinematic macro, shallow depth of field, rich <habitat> color grade, no text, no watermark". ### 5. Animate each beat `generate_video` with: - `model='i2v'`, `input_urls=[<keyframe asset_id>]` (1st entry is the start frame) - `orientation='portrait'`, `duration` = that beat's length in seconds (3 / 4 / 3 by default) - `name` = "<animal> — beat <n> motion", `entity_type='storyboard_videos'` - Prompt = the motion line for that beat (camera move + subject action), plus the style line. Keep it to one continuous action; do not ask the model to re-cut between shots. ### 6. Voiceover `generate_tts` with: - `model='tts.design'` (voice design — pass `voice_instruction`, no reference audio, no `mode` param) - `voice_instruction` = "deep, calm male narrator, slow