Automotive Campaign photography Global

Agentic campaign photography, built for a global automaker.

A premium global automaker with €65.5B in annual revenue now produces campaign-ready vehicle photography for new scenes and new markets, without booking a reshoot.

95%Brand-compliant images ready to use
100%Original vehicle preserved across every scene
More creative options from the same vehicle
01 · Client

One of the world’s largest premium automakers.

A German manufacturer of premium passenger vehicles, selling in more than 100 markets and delivering around 1.64 million vehicles a year.

IndustryPremium automotive
Employees> 50,000+
Annual revenue>$10B
Vehicles delivered~1.64M / year
Function in scopeCampaign photography
02 · The problem

Where the cost of a campaign was going.

One campaign needs dozens of angle and scene variations for a single vehicle. Every variation started as a physical shoot, and the cost showed up in 4 places.

01

New scenes meant more production

A different location or angle required another vehicle setup, crew, shoot and post-production.

02

Global campaigns multiplied the workload

Different roads, signs, number plates and driving sides meant creating and reviewing new versions for each market.

03

Existing photography had limited reuse

Once a shoot was complete, the images were tied to their original scenes. A new brief meant another production cycle.

04

Brand accuracy left no room for errors

The vehicle, branding and market details had to be right. A wrong badge, proportion or number plate could make the image unusable.

03 · Native limits

What a single prompt could not do.

The obvious approach is to generate the vehicle and scene together. But the model can alter the vehicle’s design and brand details. The real vehicle had to stay unchanged while the scene around it changed. Four things had to hold on every image.

01

Keep the real vehicle

The vehicle comes from studio photography and stays exactly as shot. No generated bodywork, no drifting badges or trim.

02

Build the world around it

Clean plate, light, weather, ground, contact shadow and reflection, all matched to a vehicle that is already fixed in frame.

03

Make it true for the market

Roads, signage, number plate and driving side, correct for the region the image will run in.

04

Check it before a human does

Scored against the brand rulebook and against the real vehicle reference, with flaws flagged for review.

04 · What Lyzr built

Six stages. One locked vehicle. Three variants to choose from.

Each stage handles a specific part of the process, from preparing the scene to generating and refining the final image. The vehicle stays unchanged throughout.

Six-stage pipeline
01 · Clean plate
Removes existing vehicles and people from the reference background.
02 · Scene understanding
Reads light, weather and mood, then localises the scene for the target market.
03 · Vehicle analysis
Reads the reference vehicle’s view, camera height and horizon, then matches the scene to it.
04 · Prompt refinement
An automated critique pass tightens the instruction for light wrap, shadow and reflection.
05 · Image generation
Generates the composite with the vehicle locked in place, at the reference’s own dimensions.
06 · Detail transfer
Transfers studio-grade surface detail back onto the chosen image.
05 · Architecture

Multiple images run without slowing down individual requests.

Each image request is processed separately in the background. Large sets of images run on their own, so they never hold up someone waiting for a single image.

Image generation architecture: the frontend calls a FastAPI service whose read routes answer from MongoDB and whose write routes publish to an interactive or a batch queue. A worker unit claims each angle, masks the vehicle first, runs the pipeline and uploads, calling SAM-3, Lyzr Studio, Gemini Image and Frameshift along the way, with Frameshift returning through a signed webhook.
06 · Controls & failure handling

Five checks between a generated image and a campaign.

The risk is not a broken image. It is a convincing image with the wrong vehicle in it. Five controls exist to stop that reaching anyone.

Locked vehicle
Composited from studio photography. Geometry, paint and angle are never regenerated.
Resolved in code
Number plate, mirroring and driving side are computed in code, so no model rewrite can lose them.
Rulebook check
Every image is scored against the brand’s global visual rulebook before it counts as usable.
Flaw inspection
Outputs are compared against the canonical studio render and flaws are flagged for review.
Human selection
Three variants per angle. Nothing ships without a person choosing it.
07 · Re-imagined workflow

One studio photo. Forty finished images.

Nothing changes about the car in the frame. Everything changes about how the world around it gets built.

Before Lyzr
Eight steps, a studio and a real vehicle. Then do it again for the next one.
and again for the next angle, scene or market
New scene
Write the brief
Find a location
Book the studio
Move the vehicle
Shoot day
Select frames
Retouch
Brand review
Campaign image
After Lyzr
Three steps, no studio. The photograph you already have does the rest.
flagged
Existing photo
Build the new scene
Brand check
Pick the shot
Campaign image
Held for review
40images from one run
90 minstart to finish
0studio days
08 · Results

What changed, per campaign image.

The same campaign image, made the old way and the new way.

MeasureBefore LyzrAfter Lyzr
To get a new sceneBook and run a shootBuild it around photography you already have
To get 40 images40 separate setupsOne run, under 90 minutes
To add a marketShoot it again for that regionSame images, localised in the pipeline
To check the brandA person reviews each imageEvery image scored automatically, flaws flagged
The vehicle itselfPhotographed again for the sceneThe original studio photograph, reused
40Images from one run
Under 90 minStart to finish for those 40
3Build generations, prototype to production
09 · What’s next

More model lines, more markets, then a move in-house.

Adding a model line means adding reference assets, not rebuilding the pipeline. The platform is contracted to migrate into the client’s own cloud environment in a later phase.

Three rebuilds from prototype to the pipeline running today

Got a use case in mind?

Platform, engineers and governance all in. We’ll map your workflow against the same 4 tests this one had to pass.