← All projects05 / 18
Boeing · AI

Boeing Generative AI Photo Experience

A high-throughput AI photo booth for Boeing’s global trade show exhibits.

IMG_20260415_145428.jpg — booth in exhibit
ROLEPipeline design + build
TECHNOLOGIESDiffusion pipeline, Python
CLIENTBoeing
YEAR2026

Boeing wanted to explore the use of an interactive LED wall as part of an attention-grabbing trade show installation. My team proposed ‘Mirror AR’, using a depth camera to composite visitors into a recreation of the 787 Dreamliner cabin. My role involved initial prototyping to evaluate and downselect depth camera configurations, as well as the inception and implementation of a low-latency, reliable & offline generative AI photobooth pipeline responsible for enhancing the captured image. My prototyping efforts affected top-level decisions about the design and behavior of the experience, and my novel AI image enhancement pipeline was responsible for achieving the quality and fidelity demanded by the Boeing creative team.

I was responsible for designing and implementing an end-to-end generative imaging pipeline that improved visual quality while remaining fast enough for a live event environment.

The challenge

IMG_20260415_144415333_HDR.jpg — booth floor

The installation had several competing constraints:

Rather than treating this as an image generation problem, I approached it as an experience design problem: determining where AI added value without disrupting the flow of the installation.

My role

I designed and implemented the complete software pipeline, serving as the bridge between Boeing’s creative direction and the engineering required to deliver it. This included:

Key design decisions

One of the earliest decisions was recognizing that visitors are extremely sensitive to changes in their appearance.

Instead of allowing the diffusion model to regenerate the entire subject, I masked facial regions — and later most of the subject except a narrow silhouette boundary — so the AI focused on the areas where it added the most value:

This produced images that felt substantially more realistic without making visitors feel that “AI had changed them.”

Iteration process

Because many image improvements were subtle, I built a lightweight browser-based comparison tool that allowed side-by-side A/B review with an interactive slider.

This became an important part of the design review process, making it easier for Boeing’s creative team to evaluate small visual differences and provide actionable feedback during iteration.

The comparators below recreate that review tool: pick a pipeline stage for each side, sweep the handle, and scroll to zoom into details — both sides pan and zoom together.

00 zedComp
03 qwen
scroll to zoom · drag handle to compare
Group — pipeline stages
tmp_b78ed835-89d8-4163-acea-de946b2e2b87.png — composite sample
01 zedComp
03 qwen
scroll to zoom · drag handle to compare
Solo — pipeline stages
00 zedComp
01 qwen_noFace
scroll to zoom · drag handle to compare
Crowd — pipeline stages

Internally, I benchmarked multiple combinations of models, hardware, and pipeline configurations, balancing image quality against runtime until the system consistently met the project’s performance target.

QWE_fp8.png — pipeline benchmark
QWE_fp8_LORAx1.png — pipeline benchmark
QWE_bf16_LORAx2.png — pipeline benchmark

Outcome

IMG_20260415_144543211_HDR.jpg — visitors at the booth

The final installation became a featured attraction within Boeing’s trade show exhibit.

Across two major aviation expos, approximately 2,000 visitors participated in the experience, receiving AI-enhanced photos via email after interacting with the booth. The installation generated visitor contact information for Boeing while creating a memorable, high-quality interactive experience.

Following the project’s success, Boeing adopted the installation as a repeatable component of its trade show presence.

IMG_20260722_100033.jpg — final composite sample
← All projectsNext: MR Materials Configurator →