Boosting E‑commerce Video Production with a Single‑Image‑Driven Workflow: A Robot‑Dog Case Study

In e‑commerce product video creation, balancing visual consistency across multiple camera angles with production efficiency is a core challenge. This case study uses a 40‑second living‑room scene for a robot‑dog product, aiming to cover panoramic, medium, and close‑up shots while keeping a unified perspective, lighting, and material rendering.

Initial Materials and Goals

The workflow starts from a single, approved main‑scene image (the living‑room layout, tone, and atmosphere are fixed). The robot‑dog and human figures are not yet placed.

Goal: extend this main image to cover 70‑80% of the video’s shots, including composite of characters and product, consistent lighting, and three storyboard types: full‑scene, side‑view, and close‑up interactions.

Stage 1 – Fusion of Characters and Scene

Import the base living‑room image into Lingjing’s Infinite Canvas as the background layer.

Import a three‑view reference of the human and a front view of the robot‑dog, describing the desired composition via prompt.

Use the Lingjing‑Banana 2‑Flash/Seedream 5.0 Lite model to generate a composite image that merges the characters and product with the scene. This composite becomes the reference for all subsequent storyboard extensions.

Stage 2 – Multi‑Perspective Storyboard Extension from a Single Image

With the fused main image, no new external assets are introduced. Prompt engineering and partial re‑rendering drive the generation of each storyboard:

Storyboard 1 – Front‑view extension : Prompt “extend front angle, girl leaning her head on her knees, other content unchanged”. Generates a medium‑shot front view.

Storyboard 2 – Close‑up of robot‑dog texture : Prompt “reference this scene, close‑up of robot‑dog looking at the girl”. Produces a detailed facial‑level shot.

Storyboard 3 – Alternate human pose : Prompt “girl no longer wearing slippers, hugging the robot‑dog on the sofa, robot‑dog looking at the girl”. Follow‑up prompt creates an eye‑pupil close‑up.

Storyboard 4 – Top‑down playful shot : Prompt “extend top‑down close‑up of toy dog”. Additional video‑generation prompts add emojis behind the robot‑dog for extra visual interest.

Stage 3 – Video Generation and Output

The sequence of storyboard images is fed into Lingjing’s video generation models: KeLing V3 and Pixverse V6 (video mode) . Parameter settings include:

Panoramic shots: camera command “slow push‑in, cinematic lighting”.

Close‑up shots: camera command “slight sway, focus on robot‑dog’s eyes”.

After rendering, the clips are ordered in editing software, background audio is added, and the final video is exported.

Stage 4 – Workflow Efficiency Analysis

Reduced material dependency : The entire process requires only 1–3 finalized visual assets, eliminating multi‑camera shooting or 3D rendering.

Visual consistency assurance : All extended frames inherit the spatial data and lighting parameters of the main image, avoiding typical lighting mismatches.

Faster modification response : Adjustments to character position or product orientation are made by editing the prompt for the single base image, without re‑generating all assets.

The approach yields roughly a 40% time saving compared with a workflow that generates each shot independently.

It is best suited for products with limited features and simple narrative scripts that require frequent multi‑angle showcases.

Conclusion

The robot‑dog product video demonstrates that Lingjing’s Infinite Canvas, combined with AI image‑to‑image and video models, can significantly shorten production cycles while preserving visual quality. Teams needing high‑frequency, multi‑angle product videos can adopt this single‑image‑driven workflow as a reference.