How image AI transformed our workflow at Cabify

This is a snapshot of our workflow in a constantly evolving field. Published in September 2025, it documents our current approach to generative AI.

At the Marketing Design Team, we produce thousands of digital ads each year — which means we need a diverse library of images and videos. Before generative AI, we depended on stock galleries to fuel our campaigns. The problem? Those same assets were available to every other brand, making it tough to find visuals that truly aligned with our messages.

On average, we’d spend about 40 minutes digging through stock libraries and finding one image, and 80% of the images we found weren’t exactly what we were looking for — a bottleneck that slowed the creative process and limited originality. As a global brand, we need our communication to be coherent across all formats. While sourcing the ideal image for Spain was straightforward, finding equivalent visuals for other regions was a constant challenge. And if we needed a matching video to pair with the static asset, that was nearly impossible.

Today, that reality has completely changed.

From search to creation: our new reality

If we want a fresh set of images today, we no longer need to spend over $6,500 a year on stock gallery memberships or $20,000 on a professional photoshoot. Now, with a $60 monthly subscription and the expertise of our design team, we can generate exactly what we need.

We’ve gone from spending 40 minutes searching for images to generating basic ones in just 10. But creating a professional-quality image still requires more time—time for refinement and creativity to ensure the result feels authentic and on-brand.

We no longer depend on generic stock image banks; we create unique visuals that are consistent for each project. Adopting generative AI for image creation has revolutionized not only our timelines and budgets but, most importantly, our creative control.

So far, we’ve seen that campaigns using AI-generated images perform better than those relying on stock photos. We’re still testing it, and soon we’ll have more metrics to validate these results.

Prompt engineering: the importance of structure and using the right words

Creating compelling images with generative AI requires more than inspiration—it demands structure. We’re literally building new worlds from scratch using the right words.

The 8-step technical structure

1. Main subject and description: Role, age, clothing, and distinctive features
2. Action or pose: What the subject is doing and their body language
3. Secondary object and description: Vehicles, objects—specifying model, color, interior/exterior
4. Environment and context: Specific city, time of day, atmosphere
5. Style and lighting: Realistic/cinematic, daylight, soft shadows
6. Mood: Confidence, calm, comfort—the feelings we want to evoke
7. Technical parameters: Camera specifications, lens, depth of field, quality
8. Aspect ratio: Square image 1:1 , widescreen 16:9, or vertical 9:16

Camera presets that work

• For car interiors or portraits: shot on Canon EOS 5D Mark IV, 50mm f/1.4, shallow depth of field, bokeh Result: Natural close-up with soft blur and excellent low-light performance.

• For exterior street photography: captured with Sony A7R IV, 35mm f/1.8, realistic street photography, handheld feel Result: Wider frame including urban context without losing realism.

Brand coherence: our non-negotiable principles

Every image we generate must align with Cabify’s reality and communicate the right messages:

• Safety concept: The seatbelt must always be worn from the right angle from drivers.
• Realistic urban environment: Authentic streets, recognizable skylines, and accurate vehicle interiors. We meticulously check for deformations and proportions, especially window sizes.
• Consistent natural light: We use specific tokens like daylight, blue sky, and soft shadows to generate believable shadows and lighting coherence.
• Human realism: Natural skin texture, anatomically correct hands, and coherent reflections that avoid the typical AI “plastic effect.”
• Brand colors: Strategic use of purple/violet accents in clothing to maintain brand coherence.

Practical examples: prompts in action

Prompt for a Driver campaign:

Chilean man, around 35, wearing a dark blue jumper, casually standing next to a modern black Sedan, checking his smartphone. The car is parked on a clean urban street in Santiago, Chile. In the background, a concrete government building with realistic architectural details. Bright sunny day, sharp shadows. Mood: relaxed, casual, professional. Realistic street photography style, captured with Sony A7R IV, 35mm f/1.8.

Prompt for a Rider campaign:

Spanish woman, 27, with long red hair, wearing a purple shirt, smiling at the camera from the back seat of a premium black car. The window creates a blurred vignette focusing on her face. Cinematic style, vibrant sunlight, slightly off-center composition. Mood: cheerful, natural, confident. Shot on Canon EOS 5D Mark IV, 50mm f/1.4, shallow depth of field, beautiful bokeh.

Prompt for a Logistic campaign:

Businesswoman, 40, dark hair, professional violet polo shirt, standing confidently inside a modern logistics warehouse. Shelves, boxes, and industrial elements are slightly blurred in the background. Natural daylight, soft shadows. Mood: professional, efficient, trustworthy. Photorealistic style, Canon EOS 5D Mark IV, 50mm f/1.4.

Prompt for a B2B campaign:

Argentinian man, 30, professional blue shirt, working on a laptop at a minimalist white table. In the background, a woman, 35, violet blazer, slightly out of focus. Bright modern office with Buenos Aires skyline visible. Mood: collaborative, focused, professional. Photorealistic style, Sony A7R IV, 35mm f/1.8.

Efficiency & sustainability in practice

At Cabify, sustainability is a core value that we also apply to our creative processes. While we don’t control the power source of AI servers, we optimize our workflows to consume less processing time, which in turn reduces GPU energy consumption.

Our sustainable workflow

• Plan and detail your prompt: We think carefully before executing. A well-structured prompt avoids unnecessary regenerations.

• Parameter intelligence:

  • Image quality: We use lower qualities for exploration and reserve maximum quality for rendering (only for final images).
  • Early cancellation: If a composition looks wrong in the first few seconds, we cancel the process to free up resources immediately.
  • Input optimization: It’s always better to write prompts in English because it eliminates the latency of a potential machine translation.

The right tool for each task

Each AI tool has its own strengths. For example, Midjourney or Gemini is great for creative exploration and complete image generation, while Photoshop AI excels at refining specific details. Since regenerating an entire image just to fix a small issue consumes unnecessary GPU power, we’ve developed a hybrid workflow to avoid this.

By choosing the most efficient tool for each stage of the process, this method not only improves the final quality but also significantly reduces unnecessary GPU usage, allowing us to effectively balance efficiency with our sustainability goals.

Rethinking the future

Integrating AI tools into our workflow has been a paradigm shift. We’ve moved from a reactive, search-driven approach to a proactive, creation-focused one. Success doesn’t come from the tools themselves—it comes from a disciplined workflow and the Design Team’s ability to apply a critical, creative eye.

The future of design isn’t about machines doing the work—it’s about humans and AI collaborating to create better, more purposeful visual communication. At Cabify, we’ve found that balance: using technology as a vehicle to amplify our creativity.