Video story lab
Generate original AI-video concepts, scripts, scene plans, narration, dialogue, and platform-native story structures from a taste profile, brief, trend, product, location, or theme. Use when the user wants ideas, short-video scripts, narrative arcs, 15-90 second plans, TikTok/Reels/Shorts concepts, campaign stories, or taste-guided creative directions.From its SKILL.md
npx -y skills add g0dam/video-studio-skills --skill video-story-labAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.2 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it
Video Story Lab
Use this skill after a taste exists, or when the user gives a creative brief that should be planned in a taste-aware way. The goal is an original story system that can feed asset design, shotboarding, render prompts, and release QA.
Workflow
- Read the chosen
taste/<slug>/taste.mdandprompt-vocabulary.mdwhen available. - Extract the user's objective: platform/provider, duration, aspect ratio, audience, tone, product/character/location constraints, language, and whether the output is for AI generation or live shooting.
- If the target provider is Seedance, normalize the deliverable aspect ratio to one of the Seedance menu options:
21:9,16:9,4:3,1:1,3:4,9:16. - Map ultra-wide
2.33:1,2.35:1, or2.39:1creative requests to Seedance21:9. Record the creative intent as "2.33:1-like ultra-wide", but use21:9as the provider ratio. - Do not invent arbitrary Seedance ratios such as
2.33:1in downstream handoff fields.
- If the target provider is Seedance, normalize the deliverable aspect ratio to one of the Seedance menu options:
- If duration is missing, offer practical defaults: 30s high-density social cut, 45s compact narrative, 60s standard arc, 90s reflective pacing. If the user wants speed, infer a default and state it.
- Inspect existing project conventions before planning:
plans/,characters/,storyboard/,shotboards/,render-packages/,audio/, and any.agents/skills/video outputs. Continue the project's path, ratio, file naming, and language instead of starting a parallel structure. - Choose a story form from
references/story-forms.md; do not default to a vague montage. For Chinese emotional or social shorts, consider起承转合or hook-build-turn-payoff before scene writing. - Run the concept/aesthetic gate from
references/story-forms.mdbefore committing to a route. A route must state visual necessity, action mechanism, contrast/irony, first-frame hook, turn/payoff, final image, renderability, and taste fit. - When the direction is not locked, produce 3-5 concrete style/concept routes. Each route must include hook, turn, payoff, visual engine, and why it fits the taste. Reject or rewrite routes that are only "setting + cute object", mood boards, or reference-name mimicry.
- Once a route is selected or obvious, write a scene plan with exact time ranges and scene purposes. Keep scene durations practical for generation or shooting; prefer 4-15 seconds per scene for short-form work, and make the timecodes add up exactly when the final duration is known.
- Include narration/dialogue only when it improves the piece. Keep lines speakable within the stated duration. For AI-video providers with unreliable sound or subtitles, mark voiceover, dialogue, music, ambience, and subtitles as post-production by default.
- Flag dependencies for the next skills: required characters, recurring places, props, visual continuity anchors, risky generated text, audio/post-production needs, and render-sensitive motion.
Output Contract
Save story outputs under the user's project folder, normally:
plans/<project-slug>-concepts.md
plans/<project-slug>-script.md
plans/<project-slug>-scene-plan.md
If the project already has a naming convention, follow it. Do not overwrite a locked plan unless the user requests replacement; use a revision suffix.
Quality Bar
- Every concept must be shootable or generatable, not just a mood.
- Every concept must need video. If the idea works just as a still poster, rewrite it until motion, timing, procedure, or a visible reversal carries the premise.
- A concept is incomplete when it is only a setting plus an appealing character or object. It needs a practical action mechanism, a legible contradiction, and a final image the viewer can remember.
- For social release, identify the comment trigger: the private truth, reversal, awkward recognition, or shared-life detail that makes viewers want to reply. Pretty atmosphere alone is not enough.
- The first 1-4 seconds must have a clear attention mechanism for short-form work.
- If an existing story/script changes materially, call out downstream assets that must be updated: storyboard text, storyboard image, provider prompts, audio plan, captions, and release copy.
- The output language should match the user's language unless they ask otherwise.
- The taste should shape decisions visibly: rhythm, shot grammar, visual motifs, sound, and text behavior.
- The plan must remain original and not recreate the analyzed creator's exact video, persona, catchphrases, or branded identity.
- List the handoff package for
video-asset-bibleandvideo-shotboard-director. - When provider constraints are known, list both the creative framing intent and provider menu ratio. For Seedance, the provider ratio must be one of
21:9,16:9,4:3,1:1,3:4,9:16.
Resources
references/story-forms.md: story forms, time structures, and output templates.
What ships with it: 2 files
7.2 KB alongside SKILL.md
agents/
- openai.yaml314 B
references/
- story-forms.md6.9 KB