# Motion Handoff > Create AI-video motion handoff prompts and keyframe grouping plans from accepted storyboard frames. Use when the user has storyboard images and needs Higgsfield Canvas, Gemini/Omni, Runway, Krea, or another video model to understand frame order, motion intent, clip stitching, and UI stability constraints. Does not generate storyboard frames; pair with videos:storyboard only when new frames are still needed. - Skill: `kennykankush/motion-handoff` (Agent Skill) - Install (CLI): `npx skillmds@latest add kennykankush/motion-handoff` - Raw SKILL.md: https://api.skillmd.com/api/skills/kennykankush/motion-handoff/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: kennykankush (https://skillmd.com/u/kennykankush) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/kennykankush/motion-handoff --- # Motion Handoff - AI Video Keyframe Director This skill turns accepted storyboard frames into motion-ready instructions for AI video tools. It is not a storyboard generator. It assumes the visual direction is mostly approved and focuses on helping a video model interpret the frames as a sequence instead of hallucinating unrelated product footage. ## Activation Use this skill for requests like: - "give me the Higgsfield prompt for these frames" - "how do I connect these storyboard images in Canvas?" - "make frame 1 to 3 animate as one clip" - "split these into AI video batches" - "write stitch-safe motion prompts" - "the video model is pasting the UI instead of animating it" Do not use this skill for first-pass narrative planning, visual style exploration, frame QA, or image generation unless the user explicitly asks for those parts too. ## Preflight Before writing handoff prompts, establish or state assumptions for: - target tool: Higgsfield Canvas, Gemini/Omni, Runway, Krea, Kling, Seedance, manual editor, or unknown - input shape: one image, start/end images, multi-keyframe node, or connected image nodes - clip duration and aspect ratio - accepted source frames and their order - whether the frames are keyframes, style references, or only visual inspiration - whether the clip must animate in, animate out, or loop - stitch strategy: overlap boundary frame, clean hard cut, crossfade, or end-card hold - stability constraints: exact UI text, logo fidelity, upright UI, no camera drift, no extra screens - what should change over time: position, scale, opacity, typing, glow, card entrance, chat bubbles, buttons, counters - what must stay locked: brand mark, product UI layout, readable labels, crop, background palette If the order or target workflow is unclear, ask one concise question before writing the final prompt. ## Core Rule When multiple images are connected to a video generation node, explicitly define them as the timeline. Use wording like: ```text Treat the connected images as sequential keyframes in this exact order, not as separate style references. Image 1 is the opening state, image 2 is the motion midpoint, and image 3 is the final state. ``` Never assume the tool will infer order correctly from upload order alone. ## Output Formats Choose the smallest useful output: - a single paste-ready prompt - a node workflow - per-shot prompt files - a batch manifest - a short handoff note with model settings and risks For node workflows, use a concrete shape: ```text Image Nodes [01, 02, 03] -> Video Generation Node Model: Duration: Aspect: 9:16 Prompt: ``` ## Prompt Anatomy A good handoff prompt has six parts: 1. Intent: name the clip's job in the launch video. 2. Keyframe order: define each image as a stage of the timeline. 3. Motion: describe only the transitions the model should create. 4. Product constraints: preserve UI, text, logos, and layout. 5. Style constraints: product-launch motion design, clean gradients, glass, spacing, typography. 6. Negative constraints: no new screens, no realistic footage, no random text, no invented logos. Example: ```text Create a polished SaaS product-launch motion-design clip using the connected images as sequential keyframes in this exact order, not as separate style references. Image 1 is the quiet opening state, Image 2 is the midpoint where the UI modules ease in and glow, and Image 3 is the final resolved product state. Animate with clean 2D/2.5D motion: soft ease-in, slight scale, opacity reveal, subtle teal glow, and controlled parallax in the background only. Preserve the product UI structure, logo, readable labels, button shapes, and 9:16 composition. Do not add new screens, fake dashboards, random paragraphs, realistic football footage, or invented brand marks. The result should feel like a premium SaaS launch motion graphic, not a phone recording or cinematic sports ad. ``` ## Stitch-Safe Sentences Add stitch guidance only when the clip will be joined to neighboring clips. Opening clip: ```text Begin from a near-empty clean stage for the first few frames so the previous clip can cut into it smoothly. ``` Middle clip: ```text Start by matching Image 1 exactly for a brief hold, then move through the sequence and end by holding Image cleanly for stitching. ``` Final clip: ```text End with a calm resolved hold on the final brand/product frame; avoid extra flourish after the last keyframe. ``` ## UI Stability Guidance AI video tools often fail when UI is too angled, too dense, or text-heavy. Prefer: - upright or nearly upright UI plates - one primary module per shot - large readable labels - stable crops with minimal camera travel - controlled opacity/scale reveals - textless or minimal-text plates when copy can be added later Avoid: - asking the model to invent typing-heavy UI - rapidly transforming a full dashboard - angled UI that must resolve into exact upright UI - start/mid/end frames with different layout systems - prompts that say "make it dynamic" without naming the exact motion ## Failure Diagnosis If the model output looks wrong, classify the failure before reprompting: - `sequence confusion`: it treated images as references, not ordered keyframes - `paste failure`: it pasted a UI image onto the background instead of integrating motion - `text drift`: readable labels mutated or became gibberish - `camera drift`: the clip became cinematic footage instead of product motion - `over-animation`: too much 3D, rotation, particles, or realism - `under-animation`: static image with only late element spawns Then write a narrower prompt that fixes that exact failure. ## What Not To Do - Do not generate new storyboard frames. - Do not manually edit images. - Do not create fake intermediate frames unless the user asks for more storyboard generation. - Do not overfit a prompt to a fixed frame number like "1 to 3" unless those are the actual selected frames. - Do not assume Higgsfield, Gemini, or any other tool is the target if the user has not said so.