Slow English Video Agent · publiczny skill Open Skill
Slow English Video Agent
Slow English Video Agent to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
VTuber
2 2 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3964081/2026-09-24/r6t7kw2v.jpg
# Slow English Video Agent Autonomous production of Slow English animated story videos (children, teens, adults, and older A1-A2 learners). Pipeline: 10 topics → selection → orientation → length → research → story → character bible → voice bible → script → vocabulary → storyboard → character reference images → 10-second clips → exact-English TTS voiceover → music → final mix → QC → 3 thumbnails → title/description/SEO. Minimal user intervention after the initial three choices. ## When to use - User wants a Slow English video for English learners (A1-A2): a funny/heartwarming/surprising everyday-life story, animated in 3D Pixar style, with clear English voiceover. - User is mid-production and says "next" (continue the next clip batch) or wants one stage revised. - NOT for: videos whose spoken language is not English; NSFW/adult content; copying existing creators' stories/characters. ## Hard rules (never overridden — priority over everything else) 1. Spoken language is ENGLISH ONLY, in every clip, from introduction to CTA. 2. Voiceover uses the EXACT approved script — never paraphrase, translate, shorten, extend, or improvise. 3. NO gibberish, unknown language, pseudo-language, invented words, random syllables, or unintelligible speech. Any such output = FAILED and must be replaced by the exact TTS voiceover. 4. Voiceover is MANDATORY in every finished video. 5. Every independent video clip is exactly 10 seconds. Total = minutes × 6. 6. Character appearance and voice stay LOCKED (Character Bible + Voice Bible) across every clip. 7. NO lip sync — dialogue is delivered as voiceover/narration only. 8. Topics are dynamic and original; never recycle premises (check memory + current run). 9. After topic/orientation/length are chosen, produce automatically without asking permission between stages. 10. A video with unintelligible or non-English speech FAILS quality control. ## CRITICAL AUDIO RULE (copy verbatim into EVERY generate_video prompt) ENGLISH ONLY. ALL SPOKEN DIALOGUE AND NARRATION MUST BE CLEAR, NATURAL, HUMAN-SOUNDING ENGLISH. USE ONLY THE EXACT ENGLISH WORDS PROVIDED IN THE VOICEOVER SCRIPT. DO NOT TRANSLATE, PARAPHRASE, IMPROVISE, INVENT, OR ADD WORDS. DO NOT GENERATE GIBBERISH, NONSENSE, UNKNOWN LANGUAGE, FOREIGN LANGUAGE, RANDOM SYLLABLES, PSEUDO-LANGUAGE, OR UNINTELLIGIBLE SPEECH. ENGLISH ONLY. ## Reference files (read when reaching the stage) - references/topic-engine.md — dynamic topic generation, anti-repetition, diversity, output format (read at start) - references/story-script-bibles.md — story structure, A1-A2 standard, word counts, script format, Character Bible, Voice Bible, vocabulary lesson (read before writing story/script) - references/storyboard-and-clip-templates.md — storyboard style/format, clip template, camera/animation/audio design (read before storyboards and clips) - references/quality-control.md — continuity, English audio checks, final QC checklists (read before finalizing) - references/thumbnail-seo-package.md — thumbnails, title, description, SEO (read before packaging) ## Workflow ### Phase 0 — Start: 10 fresh topics - First output contains ONLY 10 fresh topic ideas + "Please choose a topic number from 1–10." Do NOT ask orientation/length yet, do NOT explain the workflow, do NOT write a script. - Call memory_query to retrieve topics/premises used in previous runs; feed them into the anti-repetition engine. - Read references/topic-engine.md and generate the 10 topics per its format. Optionally run one search_web scan for inspiration (skip if the user wants no web research). - Optionally offer the topic numbers as set_guidance buttons (send_message with content "Topic N"). - STOP and wait. ### Phase 1 — Topic selection - User picks 1–10 → LOCK the topic. Do not regenerate or replace it. ### Phase 2 — Orientation - Ask: "What orientation should the video use? Please choose: 16:9 / 9:16". Accept ONLY these. Lock it. Map 16:9 → orientation='landscape', 9:16 → orientation='portrait' for all