Unit 4 Case Studies – GenAI for Industry

Product Design

Nike's AI Product Design Pipeline

Nike uses Stable Diffusion with ControlNet for rapid product prototyping, generating 500+ concept variations in hours rather than weeks. The pipeline uses negative prompts to exclude unwanted elements and inpainting for iterative refinement of designs.

Key Facts:

Design cycle reduced from 6 weeks to 3 days
Generates 500+ concept variations rapidly
ControlNet for enforcing design constraints
Negative prompts to exclude unwanted elements
Inpainting used for iterative refinement

Focus Question

How do negative prompts improve product design outputs?

They speed up rendering time They explicitly exclude unwanted visual elements They add more detail to the image They reduce memory consumption

Tip: This applies Negative Prompts and the Subject-View-Style method from Unit 4. Negative prompts guide the model away from unwanted artifacts, ensuring cleaner and more professional product renderings.

Entertainment & Film

Disney's AI Character Consistency System

Disney uses Stable Diffusion with inpainting and outpainting techniques to maintain character consistency across scenes. The system generates over 1,000 frames per scene while reducing production costs by 60% compared to traditional animation methods.

Key Facts:

60% reduction in production costs
Generates 1,000+ frames per scene
img2img with reference images for consistency
Inpainting for correcting specific frame elements
Outpainting for expanding background scenes

Focus Question

What technique maintains consistent character appearance across hundreds of frames?

Random seed generation for each frame Reference-based img2img with fixed seeds Manual touch-up for every frame Text-to-video generation only

Tip: This demonstrates img2img with inpainting/outpainting from Unit 4. By using reference images and fixed random seeds, Disney ensures the same character looks identical across hundreds of generated frames.

Fashion & Retail

Zara's Virtual Try-On with Diffusion Models

Zara uses diffusion models with inpainting for virtual try-on, reducing product returns by 25% while handling 10,000+ try-ons daily. The system precisely places garments on customer photos while preserving their appearance and body shape.

Key Facts:

25% reduction in product returns
10,000+ virtual try-ons processed daily
Inpainting for precise garment placement
Text-to-video for fabric motion simulation
Negative prompts eliminate visual artifacts

Focus Question

Why is inpainting particularly important for virtual try-on applications?

It makes the process faster It precisely fills clothing onto existing customer photos It generates entirely new images from scratch It requires less computational power

Tip: This showcases inpainting for preserving customer appearance while overlaying garments. The model modifies only the clothing region while keeping the person's face, body, and background intact—critical for realistic virtual try-on.