Input: one sharp product photo and one scene note
Output: a 5-10 second product clip for commerce
Aspect: 9:16 for TikTok/Xiaohongshu; 1:1 and 16:9 for web
Build log
1. Protect the SKU
I wrote down what could not change: silhouette, label text, logo position, color, and material. The prompt never introduces a new package or tag.
2. Pick the cheapest valid test
I start with Agnes Video 2.5 Flash at 2 credits/second. A 5-second test is 10 credits. Only clips that pass the three-frame check move to Seedance or Kling.
3. Fix camera and one variable
I keep the camera path simple and change one thing: hand motion, table surface, or light. When I changed product and scene together, the label softened.
4. Check three frames
I compare the original photo with the first, middle, and last frames. Small text is checked at full size, not in the preview thumbnail.
5. Export the platform cut
I export 9:16 first for TikTok and Xiaohongshu, then create 1:1 or 16:9 from the approved clip instead of rerendering the whole scene.
Complete execution record
Task boundary and source gate
Treat Product image to video AI as a deliverable, not a definition. The job receives one sharp product photo and one scene note and owes a 5-10 second product clip for commerce in 9:16 for TikTok/Xiaohongshu; 1:1 and 16:9 for web. Write the rejection reason first, then turn it into the first hard constraint.
Do not start until the source photo has a visible product code, a note on material and color, the approved claim, and a named target module. Check packaging damage, sticker state, lighting direction, and background rights before generation. Do not invent certifications, competitor claims, price windows, or a model using the product in an unsafe way. If the seller cannot prove it, the image does not carry it.
The imgmov route
Open Asset Library first, attach the master SKU as the locked reference, then build the module or shot around that reference. Keep offer copy, platform text, and CTA outside the visual prompt. In imgmov, the approved image becomes the source of truth and generation supplies only the requested scene, angle, or motion.
Keep platform requirements outside the model prompt. The prompt can describe a scene; a named checklist records the crop, claim, consent, and console check that make it deliverable.
Lock the variables that cannot drift
Lock product silhouette, label spelling, logo position, material, primary color, accessory count, contact shadow, and the platform safe area. Scene, props, and motion may change; these fields may not.
Put the locked fields at the top of the prompt, not at the end: Use the product photo as the locked reference and change only one scene variable. A 5-second Agnes Video clip is 10 credits; a 5-second Seedance 2.5 clip is 90 credits at 480p or 195 credits at 720p; Kling v3 Omni is 195 credits without audio and 245 credits with audio at 5 seconds. Save that prompt with the reference, ratio, and model so the next run starts from a decision instead of a guess.
Step-by-step build path
1) Protect the SKU: I wrote down what could not change: silhouette, label text, logo position, color, and material. The prompt never introduces a new package or tag. Leave one checkable artifact from this step; do not start the next until it exists. 2) Pick the cheapest valid test: I start with Agnes Video 2.5 Flash at 2 credits/second. A 5-second test is 10 credits. Only clips that pass the three-frame check move to Seedance or Kling. Leave one checkable artifact from this step; do not start the next until it exists. 3) Fix camera and one variable: I keep the camera path simple and change one thing: hand motion, table surface, or light. When I changed product and scene together, the label softened. Leave one checkable artifact from this step; do not start the next until it exists. 4) Check three frames: I compare the original photo with the first, middle, and last frames. Small text is checked at full size, not in the preview thumbnail. Leave one checkable artifact from this step; do not start the next until it exists. 5) Export the platform cut: I export 9:16 first for TikTok and Xiaohongshu, then create 1:1 or 16:9 from the approved clip instead of rerendering the whole scene. Leave one checkable artifact from this step; do not start the next until it exists.
Change one named variable per rerun: prompt, reference, camera, duration, model, ratio, or export crop. If two variables change together, a better output cannot be reused because nobody knows which fix worked.
Review gates and evidence
Review at 100%. Read small text, count buttons and parts, compare color against the source, inspect the four corners, and overlay the platform crop. A thumbnail is only a concept check, not a marketplace check.
Keep source, approved wording, rejected version, correction, credit cost, and final crop next to the asset. The record exists so the next reviewer can reproduce the decision without a chat thread.
Failure diagnosis and retry ladder
Name the failure first: label redraw, color drift, shape drift, missing accessory, wrong contact shadow, or cropped CTA. Stop prompting for more letters, reduce the variable count, return to the reference, and only then upgrade the model.
Do not upgrade on hope. A 5-second Agnes Video probe is 10 credits. If the concept passes, Seedance 2.5 costs 90 credits at 480p or 195 at 720p for 5 seconds; Kling is 195 without audio or 245 with audio. Veo enters only when a 4/6/8-second cinematic bucket is genuinely worth 80/125/165 credits. Prove hook, subject, and rhythm first, then pay to clean up motion that already worked.
Versioning and handoff
Deliver by module and ratio, not one stretched master. Name files with SKU, module, crop, version, and approval state. Keep the rejected render and the correction line in the same folder.
The folder is boring on purpose: source, approved reference, generation settings, caption file, platform cuts, QA screenshots, and one correction line. A reusable asset is boring in the right way.
Why this is not a generic answer
The imgmov advantage is the chain: Asset Library preserves the subject reference, Canvas fixes shot order and first/last frames, the workspace routes Agnes to Seedance, Kling, or Veo only after proof, and your external editor adds captions and CTA without regenerating the media.
Do not invent certifications, competitor claims, price windows, or a model using the product in an unsafe way. If the seller cannot prove it, the image does not carry it. If the request only asks what product image to video ai means, a search page is faster. This page is useful when someone must deliver a 5-10 second product clip for commerce under real constraints.
Frequently asked questions
Will the label stay readable?
Usually yes when the source photo is sharp and the motion is short. I still check small text at 100%.
The workflow is specific about its stopping point: it starts with one sharp product photo and one scene note and stops at a 5-10 second product clip for commerce. In imgmov, upload the source, lock it as a reference, set 9:16 for TikTok/Xiaohongshu; 1:1 and 16:9 for web, run one proof, and save the passing settings. If a reviewer cannot tell which version is approved, the workflow has failed even when the media looks good.
How long should the first clip be?
5-8 seconds. Agnes Video accepts 4-12 seconds; Seedance 2.5 supports the global 5-20 second range.
Do not upgrade on hope. A 5-second Agnes Video probe is 10 credits. If the concept passes, Seedance 2.5 costs 90 credits at 480p or 195 at 720p for 5 seconds; Kling is 195 without audio or 245 with audio. Veo enters only when a 4/6/8-second cinematic bucket is genuinely worth 80/125/165 credits. Prove hook, subject, and rhythm first, then pay to clean up motion that already worked.
What does a 5-second test cost?
Agnes Video is 10 credits. Seedance 2.5 is 90 credits at 480p and 195 credits at 720p.
The check is not “does it look AI-nice?” It is: Review at 100%. Read small text, count buttons and parts, compare color against the source, inspect the four corners, and overlay the platform crop. A thumbnail is only a concept check, not a marketplace check. Then the source, approved copy, rejected version, correction, and final crop stay in the same handoff folder.
Can I use the guest trial?
Yes, Agnes models have 3 watermarked previews per guest per day at 1K. They are not for commercial delivery.
Name the failure first: label redraw, color drift, shape drift, missing accessory, wrong contact shadow, or cropped CTA. Stop prompting for more letters, reduce the variable count, return to the reference, and only then upgrade the model. Change one named variable, keep the old version, and record the credit cost. If the same failure repeats, fix the reference or scope rather than asking the prompt for forgiveness.
When should I upgrade the model?
Upgrade after the scene, hook, and timing are approved. Then use Seedance 720p, Kling, or Veo only on the winning setup.
The imgmov advantage is the chain: Asset Library preserves the subject reference, Canvas fixes shot order and first/last frames, the workspace routes Agnes to Seedance, Kling, or Veo only after proof, and your external editor adds captions and CTA without regenerating the media. The handoff is reusable only when the source, reference ID, prompt, model, cost, rejection reason, and approved cut are linked. That chain is what makes the next product image to video ai task faster.