CASE STUDY 02 / AI-ENABLED CREATIVE STRATEGY + PRODUCTION

New Relic AI Image Generation

Built a custom AI image generation pipeline that transformed on-brand creative production from a manual, expertise-dependent process into a self-serve, scalable system, generating 2,400+ images and delivering a 500+ image curated library used across campaigns, events, and the full brand ecosystem.

The Challenge

When New Relic undertook a full brand transformation, one of the most crucial briefs on the table was visual expression. The existing approach leaned on generic stock imagery, which was increasingly indistinguishable from every other enterprise B2B brand. Our practitioner audience remembered when the brand had personality and play, and they weren't shy about wanting it back.

Hiring a dedicated illustrator wasn’t the answer. Bespoke illustration doesn’t scale to the volume a scrappy brand team needs to sustain. We needed a visual system that could generate highly ownable, easily abstracted imagery, consistently at scale, and on brand by default.

AI was the obvious answer. The question was how to generate something that was both special and scalable, and reliably on brand.

Early attempts made the challenges clear. Images required long, manually written prompts entered directly into ChatGPT or Midjourney. Visuals surfaced one at a time, with no brand controls in the process and no access for the broader team. The outputs weren’t perfect, but the potential was hard to ignore and the c-suite was onboard The concept had legs. What it needed next was a real workflow, something that could actually scale.

Launched with full team training and brand office hours, driving adoption from day one.

Outcomes at a Glance

2,400+ images generated. 500+ curated, approved, and available org-wide on demand.

57% faster asset turnaround, no prompt engineering required.

Brand consistency built into the backend, not dependent on the user.

The Ask


Easy

No prompt engineering. No creative gatekeeping. Anyone on the team can generate and ship.


Attractive

Dynamic, vibrant, and immediately recognizable as New Relic. Not another B2B stock photo in disguise.


On-brand

Brand colors and low-poly style baked in by default. Ownable across abstract and tangible territory.

The Process

Getting to the right output required getting to the right method first. Before a single image was production-ready, we evaluated six different approaches against the same three criteria: easy to use, visually attractive, and reliably on brand. Most tools failed on at least one count. The ones that cleared the bar still needed work to become a real system. What followed was equal parts creative direction, technical partnership, and operational design. I led creative strategy, brand governance, and program design. Our agency partner The Supervisors handled the technical build using the ChatGPT Image-1 API, and our Design Director Dan Marcus led creative direction and style refinement throughout.

Method Exploration

Finding the right tool before building the right workflow. Six approaches, one clear winner.

  • Evaluated six approaches: ChatGPT, Midjourney, Delaunay Triangulation, ChatGPT Image-1 API, a Custom Refiner Model, and Google Gemini

  • Assessed each against three criteria: ease of use, brand control, and output quality

  • Selected ChatGPT Image-1 API for its custom frontend capability, short prompts, and backend brand control

  • Ruled out off-the-shelf tools for lack of brand controls

System Design

Building brand control into the backend so users never have to think about it.

  • Partnered with an agency, The Supervisors, on technical build

  • Hardcoded brand color palette and primary prompt into the generator configuration

  • Built negative prompt to lock out off-brand styles: soft edges, gradients, heavy neon

  • Designed web app frontend for text-to-image and image-to-image input, producing up to ten variations per session

  • Developed nine creative territory themes mapped to the New Relic platform

Brand Governance

Quality isn't just a checklist. It requires a trained eye at every step.

  • Established QA review cadence with taste-level judgment at every checkpoint: technically on-brand versus what actually feels right

  • Developed quality control checklist across style, color fidelity, composition, and brand alignment

  • Uploaded approved assets to Brandfolder for company-wide self-service access

  • Built shared visual language across nine creative territory themes for consistent briefing org-wide

Rollout and Enablement

Putting the system to work, not just to shelf. Getting people using it from day one.

  • Trained creative team and key internal stakeholders on the generator

  • Hold brand office hours for ongoing creative support and real-time feedback across the org

  • Embedded imagery system into broader brand guidelines and Figma component library

  • Made library available org-wide at brand launch alongside full asset ecosystem

Methods Evaluation

Six methods evaluated against three criteria: easy to use, visually attractive, and reliably on brand. ChatGPT and Midjourney offered off-the-shelf speed but no brand controls. Delaunay Triangulation was mathematically precise but required a close input image and too much oversight. Google Gemini lacked image input capability entirely. The Custom Refiner Model showed promise but needed more work to be production-ready. ChatGPT Image-1 API, with its custom frontend, short prompts, and backend brand control, was the clear choice.

Text to Image Input

The generator’s primary mode. Users describe a subject, select tone and color emphasis, and produce up to ten variations in a single session. No prompt writing required. The brand style, color palette, and visual constraints are handled entirely in the backend.

Image to Prompt Input

The second input mode allows users to upload a reference image and either extract a prompt from it or transform it directly into the low-poly style. Users control image ratio and number of variations before generating. Useful for adapting existing assets into the brand visual language without starting from scratch.

Built In, Not Bolted On

Brand consistency isn't enforced by the user. It's built into the system. The generator configuration hardcodes the brand color palette, primary style prompt, and negative prompt directly into the backend. Every output starts from the same visual foundation regardless of who is generating or what they describe.

One Click. Multiple Variations.

Tested on the New Relic leadership team, Ashan Willy, New Relic, CEO shown here, before broader exploration and rollout. A standard photo input generates multiple stylized variations in a single click, each applying the low-poly treatment with brand color palette applied. No additional prompting required. The output is consistent, ownable, and ready to use across activations.

The Outcome

The system reduced asset turnaround time by 57%, a direct result of eliminating manual prompting, reducing revision cycles, and giving teams access to pre-approved imagery on demand.

What started as a workaround became infrastructure. The generator is now embedded in New Relic’s brand guidelines and Figma component library, with an approved asset library available company-wide.

The identity launched March 2026 as part of the full New Relic brand transformation and continues to scale across every marketing touchpoint.


Scope of Work

Strategy

  • Research

  • Executive and cross-functional alignment

Creative Ops

  • Briefing

  • Production workflow

  • Agency management

  • Budget management

  • Approval governance

  • Brand team prompt training

Design

  • Creative direction

  • Thematic development

  • Image guidelines

Enablement

  • Figma Buzz templates

  • Field and GTM toolkits

  • Google Slides templates

  • Brandfolder structure and asset distribution

  • Org-wide brand launch and office hours

Previous
Previous

New Relic Brand Transformation

Next
Next

Splunk: A ‘Legendary’ Sales Kick Off