Skip to the main content.
Login
Contact Us
Login
Contact Us
X YouTube
Responsible AI

 

Explore Our Models on Hugging Face
Try Now

Create Your First API Key
Sign Up

Need More Help?
Contact Us

Company

 

Explore Our Models on Hugging Face
Try Now

Create Your First API Key
Sign Up

Need More Help?
Contact Us

Client: Publicis’ creative agency Marcel | Aug 2024

Lidlize This!

Personalized, Online Marketing, Brand Engagement for a Global Grocery Giant 

Achieving Safe, On-Brand AI Generation at Scale
lidl header3-1
bikes lidl
1700000
+

Generations in First Three Weeks

The Client

In collaboration with Publicis’ creative agency Marcel, Lidl France harnessed Bria’s Text2Image tailored models and APIs to transform their brand identity into an interactive, personalized viral campaign

 

The Challenge

Deliver High-Fidelity, On-Brand AI Generation at Speed and Scale

Lidl France set out to redefine retail marketing by transforming everyday items—scooters, umbrellas, mugs—into symbols that reflect its iconic brand. Partnering with the global ad agency Marcel, Lidl sought to build a Gen AI solution that would enable its customers to personalize visuals in Lidl’s distinctive brand style while also ensuring safe, high-quality online outputs. 

Development Requirements:

  • High-Fidelity, On-Brand Visuals: Maintain visual integrity with Lidl’s colors (red, blue, yellow) and avoid common generative errors (e.g.: object distortions).
  • Multilingual Support: Enable text-to-image generation for English and French inputs.
  • Scalable, Low-latency Infrastructure: Support potentially high traffic with minimal latency.
  • Ethical Compliance and Content Moderation: Ensure copyright safety and uphold responsible practices in compliance with EU’s AI Act.

The Solution

A Streamlined Web App Built on Bria’s High-Quality, Compliant, and On-Brand Gen AI building blocks

Lidl and Marcel’s development teams sought a generative AI solution that could meet their stringent requirements: maintaining brand compliance, producing high-quality visuals, ensuring copyright safety, and supporting seamless online deployment. After exploring various alternatives with Stable Diffusion 1.5 to the SD XL Turbo series, Bria emerged as the only solution capable of addressing all their needs. Trained on 100% fully-licensed, commercial-grade datasets, Bria delivered on-brand, tailored, and high-fidelity outputs while adhering to strict ethical and technical standards.

 

The Development Process: BRIA Inside
Screen Shot 2024-07-29 at 5.07.36 PM (1) (1)

Tailored Model Generation

To meet Lidl’s goals, Bria adopted a structured, iterative approach to tailor a custom AI model. Serving as creative advisors to Lidl’s team, Bria’s experts helped build a dataset optimized for Lidl’s visual identity. This included carefully curating images focusing on composition, size, texture (e.g., matte vs. shiny), and aesthetic qualities. Using an agile process, the dataset was fine-tuned repeatedly to align with Lidl’s brand guidelines and artistic goals.

Leveraging Bria’s automated training pipelines, the model was retrained and tested in cycles to ensure consistent, high-quality output. This process was repeated until the model achieved precision in capturing Lidl’s signature color, texture, and composition elements.

Image credit: https://lidlize.com/en/fin

Screen Shot 2024-07-29 at 5.08.21 PM

Latency Optimization

For real-time online marketing, inference time is critical. Initial generation times of 8 seconds were reduced to a swift 2 seconds using Bria’s Fast Low-Rank Adaptation (LoRa), ensuring a frictionless user experience in high-traffic online campaigns.

Image credit: https://lidlize.com/en/fin

Screen Shot 2024-07-29 at 5.07.53 PM

Prompt Efficiency

Behind the scenes, Bria’s systems incorporated advanced large-language model (LLM) logic to generate structured prompts dynamically. This ensured uniformity across outputs while varying only the objects or specific elements, minimizing manual input. Designed for accessibility, Bria’s system removed the need for users to be prompt engineers. This democratized the creative process, empowering consumers to engage intuitively with Lidl’s branded content. Negative prompting further refined outputs by specifying elements to avoid, which helped sidestep common visual errors such as unrealistic object deformations.

Image credit: https://lidlize.com/en/fin

Screen Shot 2024-07-29 at 5.08.11 PM

API Flexibility and Language Adaptation

To improve user engagement, Bria’s API included prompt translation and leveraged DeepL Translate to support French language inputs. This reduced friction for non-English-speaking users and expanded the campaign’s accessibility.

Image credit: https://lidlize.com/en/fin

Using Bria’s model management and API endpoints, users were able to rapidly iterate on visual assets, thanks to scalable infrastructure on AWS and custom models hosted within Lidl’s control. The pipelines integrated seamlessly into Lidl’s existing systems and enabled them to support high user volumes without latency, even at peak times. Additionally, Bria’s multi-layered content moderation ensured outputs adhered to Lidl’s playful yet ethical standards, safeguarding against any offensive or inappropriate content.

Bria’s attribution technology maintained rigorous tracking and ensured all AI-generated content had a transparent origin and enabled data partners to rightful compensation to content creators. This approach aligned with Lidl’s and Bria’s commitment to ethical AI, reinforcing consumer trust and adherence to emerging regulatory frameworks such as the EU AI Act.

The Results

A Viral Marketing Campaign Increasing Customer Engagement 

With over 1.7+ million unique visuals generated by users in the first three weeks, the “Lidlize” campaign was a resounding success. The technology enabled customers to incorporate Lidl’s branded imagery into their creations, amplifying brand visibility through user-generated content shared across social media platforms. 

Credit: https://lidlize.com/en/fin

Technical Highlights and Impact
reliability

Risk-Free and Ethical Content Generation

Bria’s ethically sourced data and content moderation pipelines ensured safe, compliant outputs, strengthening consumer confidence and Lidl’s brand reputation.

equalizer

On-Brand Content Generation

Bria’s tailored LoRA Fast models allowed Lidl to maintain strict visual coherence with the brand’s aesthetic.

personalization icon

Cultural Connection

Through AI-driven personalization, Lidl was able to turn its products into cultural icons, highlighting how retail brands can leverage generative AI to drive consumer engagement beyond transactional interactions.

check-mark

Scalable, Low-Latency Infrastructure

AWS-hosted models and API integrations facilitated seamless scaling, with minimal latency even under heavy loads, supporting the campaign’s viral growth.

lidl diagram

Lidl's successful implementation showcases how Bria's infrastructure streamlines Gen AI development through rapid iteration cycles with LoRA Fast technology, AWS-native scalability, built-in content moderation, and compliance-focused design using fully licensed datasets. Their experience proves that with Bria’s models and APIs enterprises can quickly develop sophisticated Gen AI solutions that balance innovation with practical constraints around safety, scalability, and brand consistency, while maintaining high performance standards and regulatory compliance.

To explore technical documentation or start building with Bria's models and APIs, visit our Developer Platform >

Build Solutions That Redefine Marketing and eCommerce.

Build Solutions That Redefine You.

Contact Us