Product Ad Director — Photo to Video — Open Skill
Product Ad Director — Photo to Video
Product Ad Director — Photo to Video is a public Open Skill on Digen. Open it to see what this workflow creates.
SocialAd
5 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4226618/2026-10-18/sesjuyy5.jpg
--- allowed-tools: - generate_image - generate_video - generate_music - generate_tts - mix_clips - probe_media - extract_frame - extract_audio - recognize_image - recognize_video - list_models - create_document --- # Product Ad Director — Photo to Video One product photo. A polished video ad. ## When to use - The user uploads one or more product photos and wants a short video ad (Instagram/TikTok/YouTube Shorts/Facebook/LinkedIn, etc.). - They want a product showcase, problem→solution, satisfying detail/ASMR, lifestyle demo, or unboxing/reveal ad. - They want an existing product ad re-done, extended, or delivered for a specific platform, language, or duration. ## Core experience Accept: one or more product photos; optional product name, selling point, audience, language, platform, duration, and creative style; optional brand assets, website, price, offer, or call to action. Read the supplied information before asking questions. Ask one concise, bundled question only when missing information prevents a truthful or usable result. Defaults (overridden by any explicit user preference): - Vertical 9:16 (`orientation='portrait'`). - ~15-second target, adjusted to actual model capabilities. - English captions. - Premium, photorealistic commercial styling. - One finished ad. - No voiceover unless requested. - A neutral call to action without invented discounts or urgency. ## Steps ### 1. Inspect the input Use `recognize_image(image_url=<product photo asset_id>, prompt="Describe the product precisely: shape, colors, materials, packaging, label, logo, and any visible details.")` to identify the product. Separate verified facts from assumptions. Use only supplied or clearly visible product information for advertising claims. Do not infer ingredients, certifications, medical benefits, performance guarantees, or exact specifications from appearance. ### 2. Check available capabilities Call `list_models()` (and `list_models(model_type='video')`) to confirm what is actually available in this environment: supported duration, aspect ratio, reference-image controls, audio, editing, text overlays, and export options. Never invent a model, API, tool name, or supported feature. Do not force the user to choose a model when a suitable available option can be selected automatically. Respect the user's budget and approval settings. Never interpret "unlimited" or "relaxed" as free unless the selected route explicitly confirms that. Do not submit a job with an unknown cost when the user has set a strict credit limit. Use one output by default and avoid unnecessary reference-image generations. ### 3. Plan the ad Create a compact shot plan before submitting generation. Give each shot a clear purpose and one main action. Adapt shot count and pacing to the supported duration. For a 15-second ad use this flexible structure: - First 1–2 seconds: immediate visual hook featuring the product or its relevant use. - Middle: clear demonstration and one or two verified benefits. - Final section: attractive product reveal and a concise call to action. Do not spend the opening on a logo animation or a generic introduction. Deliver the visual promise made by the hook. Avoid unsupported before-and-after comparisons. ### 4. Lock product continuity Use the uploaded product image as the visual anchor. Preserve product geometry, proportions, color, packaging, label placement, cap, handle, controls, and logo wherever visible. Keep the same product across every shot. Do not invent a new package design or replace the product with a similar object. Preserve human identity, wardrobe, hand appearance, setting, lighting, and prop state across continuous scenes. Require plausible grip, contact, weight, gravity, pouring, and object movement. Avoid floating products during realistic use, interpenetration, duplicated objects, and unexplained appearance or disappearance. When exact branding cannot be preserved reliably in generated motion, prefe