Skip to content

CREATIVEAILAB

Case study: Odette Lunettes

AI-Personalized Eyewear Generator Using Stable Diffusion and Kohya SS

To celebrate the 10th anniversary of Odette Lunettes, an innovative AI tool was developed that allows users to generate custom eyewear designs based on personal input. The tool uses Stable Diffusion, enhanced with custom training via DreamBooth and Kohya SS. The project sought precision in creating unique frame designs consistent with the brand identity, combining technical experimentation, interface development, and visual AI testing. 

Step 1: Orientation and Initial Experiments

Initial tests explored standard Stable Diffusion functionalities: 

  • Text-to-image 
  • Image-to-image 
  • Inpainting 
  • Outpainting 
  • Style adjustment 


These were tested using prompts, negative prompts, and image workflows. While partially useful, results were not stable enough for production. Further investigation into custom datasets led to DreamBooth and LoRA. DreamBooth provided more control but was abandoned due to persistent library errors. 

Switched to Kohya SS (suggested by Karen Pauwels). Kohya proved more stable and modern. Training was done via LoRA, with datasets generated using Blip Captioning and enhanced with manual descriptions based on consistent variables (shape, thickness, color, pattern). 

Example: “Round glasses with a thick green leopard-patterned frame.” 

Technical settings such as learning rate scheduler, optimizer, checkpointing, and clip skip were explored via online tutorials. 

A critical error (ACCELERATE NOT FOUND) was resolved with help from Jeroen and Rob by reinstalling Kohya. Training continued with improved datasets and custom parameters. 

Initial tests were done using ComfyUI. Masks were manually drawn to align glasses with faces, due to facial variation. Two workflows (generation vs. placement) were tested and eventually separated due to technical issues.

Interface designed to reflect Odette Lunettes’ minimalist yet rebellious style. Rob Mertens enabled a web-based version for ease of use. The mockup showed a user-friendly layout aligned with brand aesthetics.

  • First model (25 photos): ~45% success 
  • Improved model (82 photos) showed better quality 
  • Optimal settings: 
  • Steps: 15–25 
  • CFG: 7–10 
  • Denoise: 1 

Simple frames (e.g., round black glasses) rendered better than extravagant designs. 

  • Collaborated with Bob Geraets for a social media contest: users generate glasses via the tool and post with #OdetteAIChallenge. The design with the most likes gets produced. 

  • Despite technical hurdles (DreamBooth errors, Kohya bugs), the result was a functional AI tool with real user potential. The process fostered deep learning in AI training, UI development, and visual consistency. 

  • Kohya + LoRA > DreamBooth for stability 
  • Structured datasets = better accuracy 
  • Keep workflows separate when integration fails 
  • Scale gradually (25 → 82 photos) 
  • Brand consistency is key 

Generative AI demands experimentation and resilience. With the right approach, it unlocks new creative potential for brands and users alike. 

AI-Generated Absurd Social Video Campaign

This case explores the use of generative AI tools to create three short videos for Odette Lunettes. The campaign used absurdity, archetypes, and visual experimentation to highlight the brand’s dual identity: Lover & Rebel. 

Initially conceived as one narrative, the project evolved into three videos with playful transformations of Belgian monuments and brand storytelling. The concept was refined through tools like Runway, Adobe Firefly, and DreamLab. 

Step 1: Concept Brainstorm & Archetypes

The concept used layered visuals with surreal effects and dynamic shots. Key ideas: 

  • Everything Eyes: Monuments with large, animated eyes 
  • Crashing Glasses: Giant frames falling onto monuments 
  • 10 Years of Glasses: A recap of top designs or dogs stealing glasses from ruins 

Absurdity and symbolism were central, guided by the Lover & Rebel archetypes. 

Runway was chosen as the main tool. Attempts with video-to-video produced distorted results due to poor camera tracking. The project shifted to image-to-video using hand-crafted Photoshop collages. Runway allowed morphing and merge animations, leading to surprising visuals.

Runway’s Generative Sessions proved effective. By combining up to 3 images with minimal prompts, clarity was maintained. A custom image bank helped keep visual consistency across the three videos. 

Three final videos were made: 

  • Format: 1080×1920 (Instagram/Facebook) 
  • Duration: 5–10 seconds 
  • Purpose: spark curiosity and position the brand as innovative 

AI output required heavy editing in Premiere Pro. Additional AI passes were used to refine transitions. 

Mentor Michael provided key feedback that shifted the project from one story to three distinct videos. Support from Immersive Lab and Odette encouraged creative risk-taking, vital to the project. YOLOv8 was briefly tested in the third video as a technical experiment.

  • Collaboration with Odette and Immersive Lab provided valuable real-world experience. A late-stage meeting with Bob Geraets required adapting the concept mid-process. Further client interaction was limited, seen as a missed opportunity. 

  • Final editing in Premiere Pro revealed AI output always needs refinement. Inconsistencies required frequent reworking via AI tools. 

    Lessons Learned 

    • Runway was the most intuitive platform 
    • Adobe Firefly & DreamLab added variety 
    • Trial and error was essential 

  • The creator started with little AI knowledge and ended with a solid artistic workflow. Initially skeptical, they now see AI as both creative partner and tool. The project boosted both technical skills and creative confidence. 

  • Embrace unpredictability: generative tools often surprise 
  • Short prompts are more effective 
  • Postproduction is always required 
  • Client feedback matters — even in experimental work 
  • Nonlinear storytelling can be powerful 

This case shows that generative AI enhances artistic exploration, especially when creativity and constraint go hand in hand. 

An AI-enhanced co-creation & try-on experience for eyewear design

Generative AI tools are rapidly gaining ground in the creative industries. The OiagaI project, developed by Ioana Oiaga as part of a TETRA research track in collaboration with Belgian eyewear brand Odette Lunettes, demonstrates how this technology can boost customer interaction, creative expression, and brand experience. 

The central objective: to develop an AI tool that allows users to generate, customize, and virtually try on their own unique eyewear designs through a web interface. This innovative platform enables co-creation with the goal of increasing engagement, discovering potential limited editions, and activating a community that aligns with the Odette Lunettes brand: stylish, edgy, playful, and empowering. 

The project was structured around research, development, testing, and a promotional motion design campaign to inform and inspire the target audience to use the tool and share their creations. 

Step 1: Conceptual development & audience targeting

The project began with an in-depth analysis of both the brand and its audience. Odette Lunettes is known for timeless eyewear with an edgy twist, made in Europe with a strong focus on sustainability. The primary audience consists of adults between the ages of 25 and 45 who are passionate about fashion, technology, and self-expression—creative individuals who are digitally engaged and interested in personalization. 

Project goals: 

  • Encourage co-creation through an AI-powered design tool 
  • Increase interactivity with a virtual try-on experience 
  • Stimulate content creation via social media 
  • Involve customers in the brand process by making them designers of their own glasses 

To determine the best technical approach, extensive research was conducted into available AI tools. The final solution combined Kohya ss, LoRA, and ComfyUI—together forming a text-to-image generation pipeline. Users type in a description and receive a unique eyewear frame based on a custom-trained dataset using existing Odette Lunettes designs. 

Why these tools? 

  • Kohya ss & LoRA: For fine-tuning models based on the brand’s eyewear designs 
  • ComfyUI: A flexible, node-based workflow builder suitable for future VR or AR integration 
  • Gradio via Hugging Face: For creating an accessible and intuitive web interface 

A custom dataset was created from Odette Lunettes’ existing models. This dataset was used to train a generative AI model while preserving consistency with the brand’s visual identity. Prompt engineering played a crucial role in managing style, detail, and avoiding imperfections. Both positive and negative prompts were tested and refined—for example: 
 Positive: “Odette Lunettes, Valentines-themed glasses” 
 Negative: “avoid weird shapes, crooked lines, multiple holes” 

The tool was made accessible through a simple and clear web interface, offering a smooth user journey: describe – generate – try on – share. The virtual try-on functionality allowed users to see their custom eyewear designs in real time, lowering the threshold for experimentation and feedback. 

Extra features: 

  • Responsive UI/UX tested with target users 
  • VR integration prepared for future expansion 
  • Social sharing options to encourage viral engagement 

A motion design and promotional video were produced to complement the tool and explain its use. More than a how-to, the video conveys the emotional and brand-driven story behind the experience—highlighting creativity, empowerment, and the joy of seeing your own designs come to life. 

Key video elements: 

  • A playful and stylish visual tone 
  • Smooth transitions between AI-generated imagery and real-life footage 
  • Consistency with the Odette Lunettes brand: neutral base tones with bold accents, sans-serif typography, and a confident tone-of-voice 

  • The tool underwent user testing throughout the project. In the second phase, the focus shifted to optimizing UX, fine-tuning prompt structure, and improving output quality. The result: a tool that’s user-friendly, technically stable, and visually aligned with the Odette Lunettes brand voice. 

AI-Animated Social Media Campaign for Odette Lunettes

For this project, Odette Lunettes aimed to inspire style-conscious adults with a series of short and powerful videos that radiate passion and rebellion. Using AI tools such as Midjourney and Google Whisk, dynamic visuals were created to emphasize the brand’s unique identity. The social media series combines a love for small details with bold eyewear frames. The project demonstrates how AI-animated videos can bring change to media production, both in terms of efficiency and creativity.

Step 1: Goal

The primary goal of the project was to strengthen the brand identity of Odette Lunettes through innovative techniques. The videos needed to highlight the duality of the brand—its ‘Lover vs. Rebel’ positioning—while also capturing the attention of the target audience in a consistent and recognizable way.

The communication challenge was to create dynamic short-form video content that visually expressed the archetypes of Lover and Rebel. Ensuring stylistic consistency across the campaign was an essential requirement. The design challenge was therefore twofold: developing a visual language that matched the brand’s identity, while also working with the limitations and unpredictability of AI-generated imagery.

The project followed a structured process that spanned several weeks. In the early stages, research was conducted into storytelling strategies for social media campaigns, as well as into the creative potential of AI tools in video production. During weeks 34 and 35, the focus shifted to concept definition: creating a mood board, setting the atmosphere, scripting voice-over texts, and developing a storyboard. Prompts were tested in Midjourney, and raw results were generated and refined. In week 36, the first jury presentation took place, followed by feedback and self-reflection

The creative process relied on a variety of AI and digital tools. Visual concepts were developed using Midjourney and Google Whisk, which allowed for the generation of experimental imagery. The text-to-speech tool Eleven Labs was used to produce voice-overs, while Perplexity served as a sparring partner for brainstorming and refining ideas. These tools enabled the translation of abstract concepts into tangible visuals and animations.

The production phase brought together several elements: image generation, editing, sound design, color grading, pacing, and music selection. Feedback was integrated into the workflow at multiple points, leading to adjustments and optimization of the final output. The process revealed the importance of balancing AI-generated material with traditional post-production skills in order to create a polished and engaging result.

The project resulted in a series of dynamic short videos that captured the essence of Odette Lunettes. Storyboards, brainstorming sessions, and test renders played a crucial role in shaping the final visual concepts. By experimenting with multiple iterations in Midjourney and Whisk, Sarah was able to select and refine the most effective directions. The final outputs demonstrated consistency in both mood and message, while still leaving room for experimental visual play.

The project highlighted the potential of AI as a creative tool, but also underscored the importance of maintaining a balance between AI-driven processes and human creativity. Sarah reflected on the ethical and sustainable aspects of AI use, emphasizing a human-first approach in which technology remains in service of storytelling. She concluded that AI can be a powerful creative partner when used critically and responsibly.

  • AI is powerful as assisting tool
  • Balance between AI vs human creativity
  • Ethics and sustainability
  • Human first approach

By combining traditional design methods with innovative AI techniques, the project sharpened the maker’s creative and critical thinking skills. The results were satisfying, yet Sarah maintained a neutral stance toward the long-term future of AI in media production. She acknowledged both the opportunities and the limitations of the technology. The project ultimately confirmed that AI tools can be integrated meaningfully into creative workflows, provided they are guided by strong concepts and human judgment.

  • Traditional techniques vs innovative technology
  • Skills and critical thinking are sharpened
  • Neutrality against the future of AI
  • Happy with the result

 

Takeaways for your projects

AI can be a powerful enabler, but it should remain a tool in service of human creativity. Striking a balance between AI and traditional production methods ensures higher quality results. Feedback and iteration are essential when working with unpredictable AI-generated content. Ethical reflection and a human-first approach help guide responsible AI use. Finally, consistency with brand identity must remain central when experimenting with emerging technologies.

Takeaways for your projects

These projects prove that AI-powered co-creation is not just technically feasible—it’s also a powerful way to merge branding, customer engagement, and creative expression. 

What we learned: 

  • Let AI work both creatively and strategically: Use it for both content generation and user interaction 
  • Ensure brand alignment: AI output should always reflect brand aesthetics and values 
  • Use playful interfaces to drive engagement: Creative tools can be fun and functional 
  • Integrate AI into your full marketing ecosystem: The motion video isn’t an add-on—it’s a narrative extension of the tool and the brand 

Ultimately, generative AI isn’t the end goal—it’s the medium. A creative accelerator that puts the user at the heart of the design process.