Claude skills that drive the Advibly MCP to generate ad creative end to end: on-brand images, UGC-style video ads, talking-actor videos, and carousels.
Each folder in this repo is one self-contained skill:
skills/
└── <skill-name>/
├── SKILL.md # the skill itself
└── references/ # supporting docs the skill loads as needed
| Skill | What it does |
|---|---|
ugc-ads |
Generates a complete multi-shot UGC video ad from a brand plus an ad angle: a realistic AI creator image (gpt-image-2), a 5-shot direct response script, a user-approved storyboard of start frames, one vertical clip per shot animated from its frame with native spoken dialogue (Gemini Omni Flash), then assembles the clips into a finished ad. Five preset angles: testimonial, car/on-the-go, unboxing, lifestyle demo, problem-solution. |
vox-explainer |
Turns a brand plus one angle into a finished Vox-style paper-collage explainer ad, directed like a real short: picks a narrative arc and writes an approved beat map (hook in 3s, two shots per beat, a cut every 4 to 6 seconds, varied camera moves, rich per-shot element motion), lets the user pick a visual theme and image model by eye from bake-offs (american-retro, swiss-modern, punk-zine, paper-craft-cream, and more), renders one richly layered collage poster per shot with headlines baked in (nano-banana-2 by default for the paper texture, gpt-image-2 for type-heavy posters) and the real product photo composited photoreal, animates each poster into living-collage motion with SFX-only audio (Gemini Omni Flash; Seedance for end-frame reveals and true assemble-from-empty builds, Kling for real people), reaches the dramatic looks (pieces flying in and assembling, confetti, impact shake, whip) through motion prompts alone with no local scripts, stitches the shots into one spot, then narrates it with advibly_generate_voiceover and scores it with advibly_generate_music, mixing voice over sidechain-ducked music and paper foley with ffmpeg. |
claymation-ad |
Turns a brand plus one product into a finished Aardman-style stop-motion claymation ad: a locked cast-and-continuity sheet plus an approved 8-beat narrated story (setup, inciting moment, social validation, quiet despair, clay infographic, discovery, transformation, resolution), one hand-sculpted plasticine storyboard still per beat (gpt-image-2, generated sequentially so the character holds), each still animated into smooth clay motion with Gemini Omni Flash by default and Seedance 2.0 as an approved secondary fallback (SFX-only), stitched into one spot, then narrated with advibly_generate_voiceover and scored with advibly_generate_music, mixed with ffmpeg (warm storyteller over ducked music and clay foley). An explicitly requested video model always wins. The product is re-sculpted as a matte clay prop, never composited photoreal. Full 8-beat (~80s) or a 5-beat short (~50s). |
pixar-style-ad |
Turns a brand and product into a vertical, original feature-film 3D animated ad: an approved cast and 4-beat micro-story (anthropomorphized problem hook, product reveal, friendly mechanism mascot scene, product CTA), gpt-image-2 storyboard stills generated sequentially for continuity, then Gemini Omni Flash image-to-video clips by default, with Seedance 2.0 as a secondary fallback. An explicitly requested model always wins. User shorthand such as "Pixar-style" is translated into an original warm, expressive 3D-animation direction. |
stickman-animation |
Turns a brand into a finished 2D stick-figure comic ad. Locks the STYLE (flat black-outline stick figures on a pure white void, uniform linework, zero shading, snappy limited animation, comic-book VFX, a two-accent color system: the brand's primary color for the product and its energy, gray for an optional problem element) and lets the agent invent a fresh STORY per brief (a problem-to-solution pitch, a one-joke gag, a visual metaphor, a running gag, a slice-of-life; any beat count). Approves an original concept and beat list, renders one flat still per beat (gpt-image-2, anchored on the first still as a style plate), animates each into snappy limited-animation motion that preserves the flat linework (Gemini Omni Flash by default, Seedance 2.0 as an approved fallback, SFX-only), then applies the editor's On Twos effect through advibly_render_composition while mixing per-beat narration and music. The product is redrawn as a flat 2D prop, never composited photoreal. An anthropomorphized-problem device (a gremlin, a blob, a "mood cloud") is one optional tool, not a template. |
collage-motion |
Decode-then-animate pipeline for halftone paper-collage and stop-motion-graphic ads. Reverse-engineers a reference image into a field-editable JSON spec, generates on-brand stills with gpt-image-2 (store products locked via catalog photo references), then animates them into a default 4-scene set of 8s assemble-from-empty clips with Gemini Omni Flash (empty color field, cut-out pieces slide in and snap into place, native audio). The final composition enables the editor's On Twos effect by default. Labels are burned in at generation, faithful to the decoded color field by default. |
explainer-videos |
Turns a brand topic or product angle into a narrated animated explainer in one of ten visual styles. Builds an approved beat map, generates styled keyframes, animates each as a recommended 4 to 6-second shot, then composes voiceover, music, and scene audio into the finished video. Gemini Omni Flash is the default motion model, with Seedance used when an exact end-frame landing is required. |
video-restyle |
Applies a complete visual style to an EXISTING video (talking head, UGC clip) while keeping the subject's identity, expressions, lip-sync, and original audio. Analyzes the source with advibly_analyze_video (Gemini transcript + cut plan), cuts it locally into 3-10s segments at sentence boundaries, restyles each segment with Gemini Omni Flash video-to-video against one of six templates (podcast-pop sticker-cutout with rotating bold backgrounds, watercolor-wash, anime-manga, newspaper-print, notebook-doodle, neon-vaporwave), stitches the segments back together, then remuxes the original audio track over the result with ffmpeg. Captions are baked per segment; the per-segment look rotation makes the output read as a deliberately edited multi-look cut. |
First, connect the Advibly MCP to Claude: advibly.com/mcp-setup.
Then install a skill. npx skills works for any supported agent (Claude Code, Claude Desktop, Cursor, and more) and pulls straight from this GitHub repo:
ugc-ads
npx skills add advibly/skills -s ugc-adsvox-explainer
npx skills add advibly/skills -s vox-explainerclaymation-ad
npx skills add advibly/skills -s claymation-adpixar-style-ad
npx skills add advibly/skills -s pixar-style-adstickman-animation
npx skills add advibly/skills -s stickman-animationcollage-motion
npx skills add advibly/skills -s collage-motionvideo-restyle
npx skills add advibly/skills -s video-restyleexplainer-videos
npx skills add advibly/skills -s explainer-videosInstall all skills at once:
npx skills add advibly/skillsPrefer to install by hand?
- Claude Desktop / Cowork: Settings → Skills → Install, and select the skill's folder.
- Claude Code: clone the repo and copy the skill folder into
~/.claude/skills/:git clone https://github.com/advibly/skills.git cp -R skills/ugc-ads ~/.claude/skills/ cp -R skills/vox-explainer ~/.claude/skills/ cp -R skills/claymation-ad ~/.claude/skills/ cp -R skills/pixar-style-ad ~/.claude/skills/ cp -R skills/stickman-animation ~/.claude/skills/ cp -R skills/collage-motion ~/.claude/skills/ cp -R skills/explainer-videos ~/.claude/skills/ cp -R skills/video-restyle ~/.claude/skills/
Then ask Claude for what you want, e.g. Make a UGC ad for my brand. Angle: testimonial. or Reverse-engineer this collage reference and animate it.
- An Advibly account with an onboarded brand (advibly.com/onboarding)
- The Advibly MCP connected to Claude
- Advibly credits (generation tools spend the account's credits)