Jack scroll asset prompts
Skill radustefandumitru/overdrive/skills/jack-scroll-asset-prompts
Generate coordinated start-frame, end-frame, and video-transition prompts for scroll-stopping website assets. Use for assembled-to-exploded product visuals, before/after transformations, 3D hero assets, image/video generation prompts, Nano Banana, Runway, Kling, Pika, Higgsfield, Veo, or any AI visual workflow intended for scroll-driven websites. Public-safe Jack Roberts inspired workflow.From its SKILL.md
npx -y skills add radustefandumitru/overdrive --skill jack-scroll-asset-promptsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 18 stars18 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
3.8 KB, 690 tokens by cl100k_base, as published. Nobody here has run it
Jack Scroll Asset Prompts
Use this skill to create a coordinated visual prompt set for a scroll-driven website: a clean starting image, a transformed ending image, and a video transition between them.
Source attribution: public-safe synthesis inspired by Jack Roberts' public teaching on AI-generated scroll-stopping website assets. Credit Jack Roberts and link https://www.youtube.com/watch?v=TZUTe7s11-I when referencing the method publicly.
Core Decision
First decide the asset type:
- Product deconstruction: assembled object -> exploded or deconstructed view.
- Service transformation: before state -> after state.
- Brand metaphor: simple hero symbol -> richer final arrangement.
Use product deconstruction for devices, vehicles, shoes, bags, watches, furniture, packaged goods, and objects with layers/components. Use service transformation for cleaning, renovation, moving, logistics, health, repair, or outcome-driven services.
Prompt Set
Produce three prompts:
-
Start frame.
- Single subject, centered, clean background, stable camera angle, premium lighting, generous margins, no text.
- Preserve real logo/product details only if the user has rights and provided the asset.
- Use the same aspect ratio intended for the website, usually 16:9 for desktop hero work.
-
End frame.
- Same subject and camera logic.
- For products: separated components should remain orderly, readable, and spatially related to the original silhouette.
- For services: transformed result should be unmistakable but still the same subject.
- Avoid chaos at the edges; keep enough empty space for cropping and responsive layouts.
-
Transition/video prompt.
- Describe start frame and end frame precisely.
- Ask for smooth deliberate motion, consistent lighting, stable background, and readable object silhouette.
- Include hold moments at start/end so the result can be scrubbed by scroll.
- Request no text, UI, arrows, labels, watermarks, or unexpected camera movement unless the design brief needs it.
Quality Rules
- Generate or request multiple candidates when possible. The final scroll section is only as good as the start and end frames.
- Prefer high resolution over speed for final assets. Low-resolution assets become obvious in full-viewport scroll sections.
- Keep backgrounds compatible with the site background so edges disappear in the implementation.
- The start and end images should share angle, lens feel, lighting, and material treatment.
- Do not promise perfect brand accuracy from image models. Verify logos and product details manually.
- Avoid unrealistic internal components unless the user wants a fantastical metaphor. Real products should feel mechanically plausible.
Output Format
Return:
- a short concept summary
- Prompt A: start frame
- Prompt B: end frame
- Prompt C: transition/video
- optional variants: reverse, loop, slow cinematic, dark-background, white-background
- recommended generation settings: aspect ratio, resolution, number of variations, and asset naming
Routing
- Use
bananaor native image tools when actually generating images. - Use
jack-scroll-3d-sitesafter the user has a video file or frame sequence. - Use
fluid-animationsandemil-animation-polishwhen the transition or scrub behavior needs to feel more natural. - Use
brandkitorimagegen-frontend-webif the brand/visual world is not defined yet.
What ships with it: 1 file
237 B alongside SKILL.md
agents/
- openai.yaml237 B