VectraSync Let’s talk video ↗
APPLIED AI RESEARCHAN INITIATIVE BY VECTRASYNC

NeoLab↗

Better video.
Through better questions.

Exploring how generative images and video can become more reliable tools for commerce, starting with the needs of cross-border sellers.

INITIAL RESEARCH AGENDA
OCTOBER 2026
01 / OUR FOCUSSMALL EXPERIMENTS. PRACTICAL QUESTIONS.

From a striking image
to a useful asset.

NeoLab is VectraSync’s applied research initiative. The programme below is our starting agenda: proposed experiments and evaluation methods, with findings to follow as work is completed.

N°01
CONTROLLABLE VIDEO

Motion that keeps
the product in focus.

Which shot constraints help short generated clips remain usable?

Proposed experiment
Test short image-to-video sequences with fixed product references, comparing restrained camera moves and scene motion.
What we will evaluate
Product drift, temporal consistency, label stability, shot adherence and usable clip length.
Planned output
A compact shot-design guide and a failure catalogue to inform production choices.
PLANNED
N°02
PRODUCT FIDELITY

Same product.
New surroundings.

How can a generated scene preserve the identity of the item being sold?

Proposed experiment
Compare reference-guided generation and compositing on a small, permissioned set of product images, varying scene and lighting.
What we will evaluate
Shape, colour, packaging text, material appearance and the amount of manual correction needed.
Planned output
A repeatable review rubric and a visual record of common failure modes.
PLANNED
N°03
CROSS-MARKET CREATIVE

A local point of view.
A consistent brand.

What should change when the same creative travels to a new market?

Proposed experiment
Adapt one approved concept across selected European, North American and Southeast Asian market briefs, changing composition, copy and channel format.
What we will evaluate
Brand consistency, text legibility, meaning, cultural fit and reviewer agreement. Market performance requires a separate live test.
Planned output
An adaptation checklist with annotated examples and local-review notes.
PLANNED
02 / RESEARCH APPROACH

Keep the experiments
small. Make the
learning reusable.

Read the initial evaluation protocol ↗

  1. Define the task

    Use owned, synthetic or permissioned assets. Set a narrow question and acceptance criteria before generation.

  2. Record the conditions

    Track model versions, prompts, references, generation settings, retries and resource use.

  3. Review the output

    Combine a structured human rubric with simple technical checks. Preserve failures alongside selected examples.

  4. Share what holds up

    Publish findings only after evaluation, with limitations and reproduction notes where permissions allow.

BUILD WITH US

A practical question
worth exploring?

We welcome conversations with brands, researchers and technology teams interested in reliable visual generation for commerce.

Talk to NeoLab ↗NeoLab is an initiative of VectraSync Limited, not a separate institution. No external affiliation is implied.