Kling is a key creator option for text/image-to-video and short scene generation. This lab tests it where product truth matters: one real product image, one simple action, one honest edit. Refresh model names and interface at lab time.

Prerequisites

• One approved product or character image (front, clean, accurate — your 59.4 hero reference).
• A two-shot plan: shot one shows the product at rest or in context; shot two shows a second angle or one simple action.
• A 59.A test card with the shared brief.

Workshop path

1. Plan one action + one camera move. Examples: "locked close-up, steam rises"; "slow push-in on label." Never compound moves. 2. Generate short candidates from the product image. Save settings, versions, time, cost. Keep clips to a few seconds each. 3. Select on fidelity first, beauty second. Freeze-frame test: label wording and position, shape, color, material, proportions. Reject beautiful drift. 4. Create the second angle. Same product, same light grammar, new framing (detail ↔ context). Carry the reference forward. 5. Run the product test on screen: labels, shape, color, hands/interactions, material behavior (liquid, fabric, steam, glass). A clip that quietly changes the product is not ready — re-generate constrained or cut it. 6. Cut a 10–20 second edit. Hook frame → proof shot(s) → one-line end card. Hand-corrected captions, CTA clear of safe-area UI, phone check muted then with sound.

What to record

• Input image name, settings per version, winner filenames, cost/time.
• The five-point product test results (label/shape/color/hands/material) with frame timestamps.
• 59.A scores for both Kling and your primary on this job, plus the decision: which tool owns product shots next time, and which owns the fallback?

Troubleshooting product shots

| Symptom | Fix |

|---|---|

| Label respells mid-shot | Shorten clip; lock camera; re-anchor to the hero still; overlay real label text in edit rather than generating it |

| Color shifts (crust, packaging) | Pin style/location reference; reduce motion intensity; compare freeze frames side by side before accepting |

| Hands pass through objects | Simplify to one hand, one slow action; cut around the contact frame; prefer a detail insert over a fake grip |

| Second angle looks like a different product | Carry the same reference file forward; match light direction and lens distance explicitly in the direction |

Worked mini-example

Shot one: locked close-up of the actual loaf, steam, morning counter — fidelity over flair. Shot two: medium angle, same loaf on the shelf beside its price label (real label photographed, not generated). The edit runs hook frame (steam close-up + "Baked at 6 AM" overlay) → shelf proof → end card with hours. Total 12–18 seconds. If either product frame drifts, the honest move is a constrained re-generation or a real photo insert — never shipping the pretty lie.

Scoring Kling against your primary (write it down)

Use the 59.A 11-dimension card on both systems for this exact product job, then decide out loud: which environment owns product shots next, which owns mood/B-roll, and what single constraint decided it (fidelity? speed? cost? availability in your region?). A typical honest outcome: "Kling owns the hero product motion because the label held across three candidates at lower cost; my primary keeps the fallback because its extension editing is faster." Record model/surface names, plan tier, date, and region — all four change, and next quarter's rematch starts from this card rather than from memory. No invented results: any dimension you did not test stays marked "not tested."

Verify fast

Pause the edit on the product's label. Can you read it — exact wording, exact placement? Do hands touch the product plausibly? If either fails, the edit fails, however pretty the motion. Note interface and pricing recheck date; Kling's models and limits change.