Lore

Kling 3.0 product-ad pipeline: product shot to finished clip

A concrete tool chain for turning one plain product image into a custom product ad clip. The worked example is AG1.

  1. Inputs: a white-background product image and a style reference from Pinterest.
  2. Image prompt: a custom GPT image prompter writes the prompt from those inputs (see Custom GPT for ad-variation prompts).
  3. Start frame: Nano Banana Pro in Higgsfield generates it (see Nano Banana Pro for product typography).
  4. End frame: Nano Banana Pro again, given the start frame plus the original product image and a prompt for a similar environment. This is prompt chaining on one image (Prompt chaining on one image for consistent frames).
  5. Motion prompt: a regular ChatGPT session receives both frames, is told the clip is for Kling 3.0, and is given the desired animation (Writing the video prompt from the reference image).
  6. Animation: Kling 3.0 in Higgsfield's video tab animates start to end.
  7. Light post-processing: a small zoom and a speed adjustment in an editor smooth the result.

Iteration

Plan for several generations per clip. Fix recurring artifacts by adding negative constraints to the motion prompt, not by redoing the frames (see closing-negative-constraints). Example: after early AG1 generations showed the green powder spilling oddly, the prompt gained an explicit instruction that the powder must not spill around the bag.

Reported results

The principle behind the chain is in Start/end frame anchoring for label-safe product animation. The author keeps the full prompts in his paid community, so only the structure is public.