Knowledge Vault 5 /94 - CVPR 2024
Today’s Pictures, Tomorrow’s Training Data: The Synergy Between Human Creativity and AI
Andrea Gagliano
< Resume Image >

Concept Graph & Resume using Claude 3 Opus | Chat GPT4o | Llama 3:

graph LR classDef getty fill:#f9d4d4, font-weight:bold, font-size:14px classDef ai fill:#d4f9d4, font-weight:bold, font-size:14px classDef trends fill:#d4d4f9, font-weight:bold, font-size:14px classDef ethics fill:#f9f9d4, font-weight:bold, font-size:14px classDef tech fill:#f9d4f9, font-weight:bold, font-size:14px Main --> A[Getty Images] Main --> B[AI in Imagery] Main --> C[Visual Trends] Main --> D[Ethical Considerations] Main --> E[Technology and Innovation] A --> A1[High-quality image provider
for diverse needs 1] A --> A2[Research program reporting
visual trends 2] A --> A3[Gettys AI model addressing
customer concerns 6] A --> A4[Model uses licensed images
for training 7] A --> A5[High-quality captions enhance
model training 8] A --> A6[Model avoids generating
branded content 9] B --> B1[Increasing AI use in
image creation 4] B --> B2[Explanation of image
generation process 12] B --> B3[Generated images show
superior quality 14] B --> B4[Method to compare visual appeal 15] B --> B5[Benchmarks for quality,
semantic adherence 16] B --> B6[AI visualizes difficult-to-capture concepts 18] C --> C1[Imagery evolution over past decade 3] C --> C2[Dataset includes contemporary concepts 13] C --> C3[Workplace imagery evolution 25] C --> C4[Energy depiction shift 26] C --> C5[Emotional representation becoming nuanced 27] C --> C6[Family depictions becoming more diverse 29] D --> D1[Legal, ethical concerns
about AI images 5] D --> D2[Aims for diverse, inclusive
image production 10] D --> D3[Emphasizes paying creators for work 11] D --> D4[Importance of compensating
content creators 23] D --> D5[Encouraging licensed AI training data 24] D --> D6[Mental health visualization changes 28] E --> E1[Human creativity meets
innovative technology 17] E --> E2[Traditional photography remains essential 19] E --> E3[Continual creation reflects
societal changes 20] E --> E4[Risk of models becoming stale 21] E --> E5[Illustrates current AI model limitations 22] E --> E6[AI empowers complex
concept visualization 30] class A,A1,A2,A3,A4,A5,A6 getty class B,B1,B2,B3,B4,B5,B6 ai class C,C1,C2,C3,C4,C5,C6 trends class D,D1,D2,D3,D4,D5,D6 ethics class E,E1,E2,E3,E4,E5,E6 tech


1.- Getty Images: A company providing high-quality images and videos, serving customers from small businesses to enterprises for various content needs.

2.- Visual GPS: Getty Images' research program that reports on visual trends, combining industry surveys and data on search and download behavior.

3.- Visual trends evolution: How imagery representing concepts like work, energy, and family has changed over the past decade.

4.- AI in imagery: The increasing use of generative AI in creating images, with 66% of Getty Images customers using it.

5.- Customer concerns: Legal, authenticity, compensation, and ethical issues surrounding the use of AI-generated images.

6.- Generative AI by Getty Images: A model built to address customer concerns, powered by NVIDIA Picasso and available on Getty Images and iStock.

7.- Licensed training data: Getty Images' model uses only licensed images (about 200 million) for training, ensuring ethical and legal use.

8.- High-quality captions and keywords: Getty Images' dataset includes manually added, reviewed, and edited captions and keywords for better model training.

9.- Brand and logo avoidance: Getty Images' model doesn't generate branded content or logos, reducing legal risks for customers.

10.- Authenticity and representation: The model aims to produce more diverse and inclusive images compared to other generators.

11.- Compensation for photographers: Getty Images emphasizes the importance of paying creators for their work and its use in AI training.

12.- Diffusion model: Explanation of how image generation models work, emphasizing the importance of original training images.

13.- Contemporary topics in training data: Getty Images' dataset includes more recent concepts like sustainability and remote work compared to web-scraped data.

14.- Quality comparison: Getty Images' generated images show higher quality and realism compared to other industry generators.

15.- Aesthetic quality scoring: A method used to compare the visual appeal of generated images across different models.

16.- FID vs CLIP plot: Industry benchmarks used to compare the quality and semantic adherence of generated images to real images.

17.- Creative magic: The combination of human creativity and innovative technology in producing impactful images.

18.- Generative AI for visualizing concepts: Using AI to create images that would be difficult or impossible to capture with a camera.

19.- Camera-shot imagery importance: Some concepts still require real people and emotions captured by traditional photography.

20.- Future of visual content: The need for continual creation of new images to reflect societal changes and provide fresh training data.

21.- Model collapse: The risk of AI models becoming stale without new, real-world training data.

22.- Four-generation example: Illustrates limitations of current AI models in generating images of concepts not well-represented in training data.

23.- Paying creators: The importance of compensating photographers and creators for their work and its use in AI training.

24.- Licensing training data: Encouraging companies to license and pay for the data used to train their AI models.

25.- Evolving work environments: How imagery depicting workplaces has changed from formal office settings to remote work and diverse teams.

26.- Changing depictions of energy: Shift in popular imagery from oil rigs to electric vehicle charging stations.

27.- Evolution of emotional representation: Movement from exaggerated, cheesy emotions to more nuanced and realistic depictions of feelings.

28.- Mental health visualization: Changes in how mental health issues are depicted visually, especially in workplace contexts.

29.- Family representation: Shift towards more diverse family depictions, including multi-generational, multi-racial, and LGBTQ+ families.

30.- Technological impact on creativity: How AI tools are enabling creators to quickly iterate and visualize complex or futuristic concepts.

Knowledge Vault built byDavid Vivancos 2024