💡 Inspired by: Reid Hastings & Creative AI Advocates — demonstrating AI as a powerful creative collaborator
AI for Creativity Writing, Design & Storytelling
AI isn't just for research — it's a creative collaborator. In this lecture, you'll co-write opinion pieces, design event posters
with Canva AI, and generate images from text descriptions. The big question: Where does AI's work end and yours begin?
Concept Discussion — 20 Min
Latent Space Diffusion & Synthetic Ontology
Generative image models do not 'collage' existing artwork. They utilize Latent Diffusion models (e.g., Rombach et al., 2022). They are parametrically trained to reverse a Markov chain of Gaussian noise, gradually interpolating variables within a high-dimensional mathematical 'latent space' to synthesize novel pixel correlations.
30%
Tasks in the creative industry that could be automated by 2030 (Goldman Sachs)
15B+
AI images created explicitly since 2022
From GANs to Diffusion
Before 2022, AI art was dominated by Generative Adversarial Networks (GANs)—two neural networks competing. They created blurry, nightmarish faces. Today's paradigm (Midjourney v6, DALL-E 3) relies on cross-attention text encoders (like CLIP) that perfectly link mathematical text vectors to visual geometry.
Prompting for Visual Vectors
Text-to-Image models require mechanical prompts, not conversational prompts. Instead of talking to the AI, you must command the 'virtual camera'.
Medium: Specify the exact substrate (e.g., "35mm analog photography", "oil on canvas", "octane render").
Lighting: Lighting controls the diffusion noise floor (e.g., "cinematic lighting, golden hour, volumetric haze").
Aesthetics/Artists: Borrowing weights from specific eras (e.g., "Bauhaus architecture", "in the style of Studio Ghibli").
The Copyright Dilemma
The New York Times is suing OpenAI, and artists are suing Stability AI. Why? Because these models were trained on copyrighted text and artwork scraped from the internet without explicit permission or payment.
Opt-out vs Opt-in: Currently, artists must explicitly opt-out of training data, rather than AI companies asking to opt-in.
The Fair Use Defense: AI companies argue that reading images to "learn" a style is transformative, similar to a human art student studying Picasso to learn cubism.
The Bangla NLP Limitation
Bangla is the 7th most spoken language globally (300M+ speakers). Yet, it suffers from linguistic inequality in AI ecosystems.
Why is Bangla AI lagging?
Large Language Models require massive amounts of high-quality, digitized text dataset (corpuses). While English has trillions of digitized tokens online, high-quality, perfectly digitized, domain-specific Bangla data is scarce.
The Opportunity: For Bangladesh to thrive in the AI era, we need linguists, programmers, and cultural experts to build native, high-quality Bangla Language Models (like the BanglaLLM initiatives currently underway locally).
🎥 More on Creative AI
Hands-On Activity — 20 Min
🖼️ Activity 1: The Design Co-Pilot
Open Canva AI or Microsoft Copilot Designer. Your faculty is hosting a massive conference. Design the main promotional poster entirely through text prompts.
Start broad:
A promotional poster for an economics conference in Dhaka.
Refine with style keywords:
A promotional poster for an economics conference in Dhaka. Add minimalist geometric shapes, use Dhaka University colors, make it look cinematic and highly detailed.
Critique the output: Did it spell the Bangla/English text correctly? (Often, image AIs fail at typography).
🪞 Activity 2: Reflection Discussion
As a class, discuss the central question:
"Where does the AI's work end and yours begin?"
If AI writes 60% and you edit 40%, whose work is it?
If AI generates the image but you wrote the prompt, are you the artist?
How should creative AI collaboration be credited in academic and professional work?
🎨
Your Takeaway: Your First AI Generated Asset
Save your best generated image or design from today. Add a short paragraph explaining the exact prompts you used to get from the initial bad draft to the final output. This is the third piece of your AI Portfolio.
