Plan AI product video shots that preserve a product’s shape, color, and label. Build references from several angles, choose safer camera moves, check each clip for visual drift, and composite the real product image where text must remain exact.
Turns product photos into AI video shots that keep the product identical: a lock sheet (fixed geometry, fixed surface, allowed motion), a reference pack of 2-4 angles passed as @image roles to Seedance 2.5 or Kling 4, a camera-risk ladder (locked frame → push-in → slide → small orbit), banned motion words, one action per 4-6 s shot, a drift check with varg video_analyze, and the real packshot composited back where text must be exact. Fixes the #1 "looks fake" tell in AI product ads: warped logos, melted labels, products that change shape or size between shots. Use when a product drifts, morphs, changes color or loses its label in AI video, when someone asks for "consistent product", "keep my logo", "product looks different in every shot", or before a batch of product videos for TikTok Shop, Meta or Amazon. Not for people and characters staying the same (use varg-consistent-characters or varg-character-design).
Run score Not tested
See how to installReview the instructions the agent will receive before you install.
AI video models redraw the product in every frame. With one photo they guess the hidden sides, and long, soft or printed products drift: wedges turn into blocks, labels melt, cans invent a back. This skill locks the product before any clip is generated and checks every clip after.
Generation runs on varg: REST https://api.varg.ai/v2 (the varg-ai skill) or the MCP server https://mcp.varg.ai/mcp.
PRODUCT: car seat gap filler, single piece
FIXED GEOMETRY: long thin wedge, ~50 cm long, ~4 cm wide at the top, tapering; seatbelt slot near one end
FIXED SURFACE: black leather-like cover, contrast stitching along the top edge, no logo
ALLOWED MOTION: held, slid into a gap, objects dropped on top; never bent, squeezed or rotated fast
NEVER: boxy, shorter, glossy, another color, a second slot
Separate geometry, surface and motion; never mix an old and a new label in the references (seedance.tv).
| Ref | What | How |
|---|---|---|
@image1 hero | front three-quarter, clean, ≥1080 px | the supplier photo, or gpt_image_2_5 to fix text |
@image2 profile | the side that shows true shape and size | nano_banana_2_1 edit of the hero: "the exact same product from the side on white" (5 credits) |
@image3 detail | label, logo or material close-up | crop of the real photo |
@image4 scale | in a hand or next to a known object | optional |
Seedance 2.5 takes up to 9 images, Kling 4 up to 7 references per video, Veo 3.1 Ingredients up to 3. Give each image one role in the prompt: "@image2 only shows the true side profile, not the background" (AuraJZ/ai-product-video).
Product lock. @image1 is the exact product, front three-quarter view; @image2 is the same product from
the side, showing its true long thin wedge shape (about 50 cm long, 4 cm wide). The product keeps this
exact shape and size: it does not become boxy, does not shorten, does not change color, stitching or the
seatbelt slot. Never re-describe it beyond this.
<scene, one simple action, slow and steady, camera holds still with a slight handheld drift>
Real footage look, no music, no text, no subtitles.
curl -s -X POST https://api.varg.ai/v2/tools/video_analyze/call -H "Authorization: Bearer $VARG_API_KEY" \
-H "Content-Type: application/json" -d '{"model":"video-analyze","video_url":"CLIP_URL",
"prompt":"Compare the product in this clip with this description: <lock sheet>. For the first, middle and last second: does shape, size, color, label and every fixed detail match? Answer PASS or FAIL per attribute with the second where it breaks."}'
About 5 credits per clip. FAIL on any fixed attribute → rerun that shot only, with a shorter duration or one step down the camera ladder. Make 3–4 takes of a hero shot and keep the most stable.
For the hero close-up, use the real packshot: a slow push-in on the photo (zoompan) or the generated scene with the packshot overlaid on a static shot:
ffmpeg -loop 1 -i packshot.png -vf "scale=1300:-2,zoompan=z='min(zoom+0.0008,1.12)':d=150:s=1080x1920:fps=30" -t 5 -pix_fmt yuv420p hero.mp4
Curved labels on moving bottles need planar tracking (Mocha) in post; avoid those shots if you can.
2026-10-28, car seat gap filler. With one photo, seedance_2_5 turned the long wedges into boxy blocks in the reveal shot. With the lock sheet, a side-profile reference made by nano_banana_2_1 (5 credits) and the lock block above, the same shot kept the long wedge shape, the color and the stitching for the whole 6 s (231 credits). Before/after: https://s3.varg.ai/files/194b94d0-58ff-4e8b-91de-4f1959d04013/file_332df2a110a0.mp4
varg-unboxing-video and varg-product-hook (shots that need the lock), product-photo-to-video, varg-consistent-characters (people, not products), varg-ai.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Plan worthwhile local tasks by checking dependencies, effects, resources, and platform boundaries, then execute the exact approved plan. It can use safe parallel work or suitable Apple-silicon backends when the project exposes them.
Turn a winning TikTok or Meta ad into a structured creative test plan. Identify what made it work, plan distinct concepts, set test budgets and review timing, and use funnel metrics to decide whether to stop, iterate, or scale each ad.
Record deferred engineering decisions and temporary workarounds in a debt registry, with context such as effort, ongoing cost, affected area, and payoff trigger. Entries receive short labels so they can be reviewed or dropped later.
Generate or compare AI agent code for LangChain, EINO, AutoGen, AgentScope, and CrewAI using framework-specific templates, API mappings, conventions, and project structures. Reference files include verification dates to help identify potentially outdated guidance.
Investigate a reproducible bug, test failure, or unexpected page behavior, then make a narrowly scoped fix. The workflow calls for a regression test before changing code, verifies the fix, and reports remaining risks and evidence while avoiding unrelated refactoring or feature work.
Build an AI agent project through six collaborative stages, from requirements and architecture to tools, prompts, code generation, and verification. Start framework-independent, then generate a runnable project for frameworks such as LangChain, EINO, AutoGen, AgentScope, or CrewAI.