--- name: product-nine-grid-video description: "Turn product images and selling points into a staged apiz.ai CLI workflow: inspect product image quality, guide missing CLI install/login when needed, optionally optimize the product image with gpt-image-2, generate multiple 3x3 nine-grid promotional images, then after user approval create a 15-second multi-reference image-to-video with st-ai/super-seed2-lite. Use when users ask for product nine-grid ads, 小红书九宫格, product promo grids, product image optimization before marketing images, product image-to-video campaigns, or natural-language APIZ image/video/audio/tool generation through the apiz CLI." --- # Product Nine Grid Video ## Core Rule Run this as a gated production workflow. Do not skip confirmation gates: 1. Inspect the input product image. 2. If the image is not clean, clear, and commercially usable, optimize it with `gpt-image-2` and show the optimized output for user confirmation. 3. After the product image is accepted, generate multiple 3x3 promotional grid options with `img2` / `gpt-image-2`. 4. Show the grid options and ask the user to pick or request changes. 5. After grid confirmation, submit the accepted product image plus accepted nine-grid image as multi-reference inputs to `st-ai/super-seed2-lite` for a 15-second image-to-video result. Prefer the `apiz` CLI for model discovery, generation, and task polling. If `apiz` is missing or not authenticated, run the preflight in [references/apiz-cli.md](references/apiz-cli.md) and guide the user through install/login before generation. ## APIZ CLI Preflight Before calling any model on a new machine: 1. Run `command -v apiz`. 2. If missing, install it: - macOS/Linux: `curl -fsSL https://apiz.ai/cli | sh` - Windows: `scoop install apiz` 3. If not logged in, ask the user for an APIZ API key or tell them to set `APIZ_API_KEY`. Do not write real API keys into skill files, logs, commits, or public files. 4. Login with `apiz auth login --api-key "$APIZ_API_KEY"` or a user-provided secret in the shell only. 5. Run `apiz account balance --json` and stop if authentication fails or balance is insufficient. Use `--json` by default on CLI commands and decide success from `status`, `result`, `error`, and `message`. ## Required Inputs Collect or infer: - Product image path or URL. - Product name/category. - Selling points, campaign copy, target audience, platform, and preferred style. - Brand constraints: required words, forbidden words, logo/text preservation, color preference. If only the image is provided, inspect it first and ask for missing marketing content before generating the nine-grid. ## Model Selection - Product optimization: find and use `openai/gpt-image-2/edit` or the available `gpt-image-2` image edit endpoint. - Nine-grid generation: find and use `openai/gpt-image-2` or the current APIZ image model that supports high-quality image generation. Use image edit with `--image-url` when the model needs the accepted product image as reference. - Video generation: find and use `st-ai/super-seed2-lite` exactly when available. If the CLI rejects that model id, search video models for `super-seed2` and use the closest lite-capable multi-reference image-to-video model only after telling the user about the substitution. - For exact command patterns, model docs, and task polling, see [references/apiz-cli.md](references/apiz-cli.md). ## Image Inspection Evaluate the product image before generation: - Clarity: sharp product edges, readable label/text, no compression artifacts. - Background: clean, not cluttered, not visually competing with the product. - Angle: commercial product angle, sufficient front/side information, no awkward crop. - Lighting: even, no harsh color cast, no blown highlights or muddy shadows. - Product integrity: no warped shape, missing parts, changed label, or incorrect texture. Classify as: - `pass`: use the original product image. - `needs-optimization`: run product image optimization first. - `blocked`: ask for a better product source image if the product cannot be identified or is too degraded to repair safely. Use [references/prompt-templates.md](references/prompt-templates.md) for audit, optimization, grid, and video prompts. ## Product Optimization When the image needs improvement: - Use `apiz models list --category image --json` and `apiz models info openai/gpt-image-2/edit --json` before the first image-edit call on a new machine. - Use `apiz generate "" --model openai/gpt-image-2/edit --image-url --wait --json` plus schema-specific `--params` as needed. - Preserve product geometry, label, logo, color, material, and package text. - Fix only commercial presentation issues: background, lighting, crop, shadow, minor dust, mild perspective. - Prefer an opaque white, light gray, or clean lifestyle-neutral background unless the user asked otherwise. - Generate 1-3 optimized options when practical. After optimization, show the output and ask for confirmation. Do not proceed to grid generation until the user approves one optimized image or explicitly approves the original. ## Nine-Grid Generation After product image approval: - Generate 2-4 distinct 3x3 grid options. - Each option must be one square image containing exactly 3 columns and 3 rows. - Make each cell readable after cropping as an independent social post tile. - Keep safe margins inside each tile. Avoid text near grid dividers. - Use Chinese copy by default for Chinese users. Do not add random English. - Keep product identity consistent across all cells. - For APIZ generation, use `apiz generate "" --model --wait --json`. Add `--image-url ` or schema-specific `--params` when using an image reference. - Prefer high quality / 2K when the model supports it because nine-grid images are text-heavy. Recommended grid structures: - `hero-selling-points`: cover, problem, product hero, 3 benefits, use scene, proof, CTA. - `before-after`: pain point, wrong solution, optimized product, steps, comparison, CTA. - `brand-kit`: product hero, color palette, ingredients/features, scenarios, packaging, CTA. - `tutorial`: cover, 6-step usage guide, mistake reminder, result/CTA. Show all grid options to the user. Ask which option to use or what to revise. Do not create video before approval. ## Video Generation After user confirms the final nine-grid: - Use `st-ai/super-seed2-lite`. - Use multi-reference image-to-video mode. - Reference images: 1. Accepted optimized product image. 2. Accepted nine-grid promotional image. 3. Any extra brand or scene reference the user provided. - Duration: `15` seconds. - Prompt should request a smooth promotional video that animates through the nine-grid cells, showcases the product, and keeps all text/product identity stable. - Use the same aspect ratio the campaign needs. Default to `9:16` for short-video/social ads unless user requests otherwise. - For vertical videos, require full-frame 9:16 composition. Do not let the model place the square nine-grid as a small horizontal strip with large white margins. - Treat the nine-grid as a design/style reference and benefit-card source, not as a literal object to slide across the screen for the whole video. - Preserve only key readable headings in video. Do not ask the model to reproduce every small paragraph from the nine-grid. - Prefer product-ad pacing: product hero, installation close-up, 2-4 benefit cards, product packshot/CTA. - Explicitly forbid giant abstract floating outlines, repeated side-scrolling galleries, warped phones, warped film, flicker, random text, and watermarks. - Use `apiz models list --category video --json`, `apiz models info st-ai/super-seed2-lite --json`, and then `apiz generate "" --model st-ai/super-seed2-lite --duration 15 --aspect-ratio 9:16 --wait --json` with schema-specific multi-reference image parameters. Poll until completion, then show the video result. ## General APIZ Natural-Language Tasks When the user asks for APIZ image, video, audio, speech, tools, or task results outside the nine-grid workflow: - Run the APIZ CLI preflight first on new machines. - List the relevant category: `apiz models list --category image --json`, `apiz models list --category video --json`, or `apiz models list --category audio --json`. - Inspect schema: `apiz models info --json`. - Load docs when unsure: `apiz models docs --lang zh`. - Generate: `apiz generate "" --model --wait --json`. - Poll async tasks: `apiz tasks get --json` or `apiz tasks wait --json`. - Summarize accessible URLs, task ID, model ID, and key parameters for the user. ## Output Protocol Keep user-facing updates compact: - After audit: state pass/fail and why. - After optimization: show options and ask for confirmation. - After nine-grid generation: show multiple options and ask for selection. - After video generation: provide the video link/file and summarize inputs used. Never claim a model result succeeded until the generation task result is complete.

product-nine-grid-video

Install this skill

Comments