Chapter 16 of 17
Generative AI in Creative Domains, Gaming and Animation
Explores AI image generation, music composition, and game and animation content such as NPC dialogue. It covers how diffusion works in practice, prompt design for visuals, and the copyright, consent and deepfake questions these tools raise.
In this chapter
16.1 Image and design generation
How it works. Most current image generators are diffusion models (Chapter 3). A text encoder turns the prompt into a vector; the model starts from random noise and removes it step by step, guided by that vector, until an image forms.
Major tools
| Tool | Developer | Notes |
|---|---|---|
| DALL·E (and successors in ChatGPT) | OpenAI | DALL·E (2021), DALL·E 2 (2022), DALL·E 3 (2023, built into ChatGPT); since 2025 ChatGPT generates images with its GPT Image model |
| Midjourney | Midjourney, Inc. | Known for artistic, stylised output; open beta from July 2022 |
| Stable Diffusion | Stability AI | Open model released August 2022; can run on local hardware |
| Adobe Firefly | Adobe | Built into Photoshop and Illustrator; trained on licensed and Adobe Stock content |
| Imagen / Gemini | Image generation inside Gemini and Google tools | |
| Canva Magic Media | Canva | Quick social graphics and presentations |
Anatomy of an image prompt
| Element | Example |
|---|---|
| Subject | "A clay water pot (matka) on a wooden stool" |
| Setting | "in a sunlit village courtyard in Rajasthan" |
| Style or medium | "watercolour illustration" / "product photograph" / "flat vector" |
| Composition | "close-up, centred, shallow depth of field" |
| Lighting and mood | "warm late-afternoon light, calm mood" |
| Colour | "earthy terracotta and indigo palette" |
| Technical | "square 1:1 format, plain background for product listing" |
Full prompt: "Product photograph of a hand-painted clay matka on a wooden stool, plain off-white background, soft studio lighting, centred, square 1:1, high detail."
Design uses: logo concepts (refined by a designer), mood boards, product mock-ups, social media graphics, book and poster concepts, storyboards, architecture and interior concepts.
Limits: text inside images may be misspelled; hands and fine detail can be wrong; consistency of a character across many images is hard without special tools; and outputs may imitate living artists' styles, which raises ethical concerns.
16.2 Music composition with AI
| Approach | What it does | Example tools |
|---|---|---|
| Text-to-music | Generates a full song, sometimes with vocals, from a text description | Suno, Udio |
| Instrumental composition | Generates background scores by genre, mood and length | AIVA, Soundraw |
| Assisted composition | Suggests melodies, chords or accompaniment inside music software | Plugins for digital audio workstations |
| Stem separation and mastering | Splits vocals and instruments; balances the final mix | Audio tools with AI features |
Example prompt: "60-second upbeat instrumental for a college fest promo video, Bollywood-pop style, dhol and synth, 120 BPM, builds to a high-energy ending."
Uses: background music for videos and podcasts, jingles, game soundtracks, quick demos for musicians, and music education.
Concerns: whether training on copyrighted recordings is lawful; imitation of real artists' voices; and the effect on income for session musicians. In June 2024 major record labels, through the Recording Industry Association of America (RIAA), sued Suno and Udio over the use of copyrighted recordings in training.
16.3 AI in gaming and animation
| Use | How generative AI helps | Human role |
|---|---|---|
| Scenario and story generation | Drafts quests, side-stories, lore and branching plots | Lead writer keeps the story coherent |
| NPC (non-player character) scripting | Writes dialogue variations and short "barks" | Writers select, edit and set the character voice |
| Dynamic NPC conversation | NPCs answer players in real time within character limits | Designers set guardrails |
| Procedural content | Generates levels, maps, items and textures | Designers tune for fun and balance |
| Concept art | Fast exploration of characters and environments | Artists refine and produce final assets |
| Animation | Text-to-motion, lip-sync to dialogue, in-betweening frames | Animators direct and polish |
| Localisation | Translates and adapts dialogue for regions | Native speakers review |
| Testing | AI agents play-test levels and find bugs | QA team analyses results |
Example: NPC dialogue prompt
You are writing dialogue for a game set in 1850s Lucknow.
Character: Rahim, an elderly chikankari embroidery seller, warm but shrewd.
Write 8 short lines (under 12 words each) he says when the player:
- walks past (2 lines), - looks at his goods (2), - haggles (2), - buys something (2).
Keep the language simple English with occasional Urdu words, no anachronisms.
16.4 Ethics, copyright and deepfakes
- Copyright in AI output. The US Copyright Office has held that material generated by AI without sufficient human creative control is not protected; in the Zarya of the Dawn comic (2023), the human-written text and arrangement were protected but the Midjourney-generated images were not. Courts in Thaler v. Perlmutter upheld that copyright requires a human author. Indian law on this question is still developing.
- Training data. Lawsuits by artists, authors, news organisations and record labels question whether training on their work without permission is lawful.
- Style imitation. Prompting "in the style of" a living artist is legal in many places but can harm that artist's market.
- Deepfakes. Realistic fake images, voices and videos of real people can be used for fraud, harassment and misinformation. India's Ministry of Electronics and IT issued advisories to online platforms in 2023 on removing deepfake content, and the IT Rules require platforms to act on such complaints.
- Disclosure. Platforms such as YouTube and Meta require or encourage labels on realistic AI-generated content.
16.5 Worked numericals
Numerical 16.1: Why diffusion works in a compressed "latent" space.
An RGB image of 1024 × 1024 pixels has 1024 × 1024 × 3 = 3,145,728 values. Stable Diffusion's encoder shrinks each side by 8 and uses 4 channels.
The model denoises 48 times fewer numbers at each step, which is why it can run on a single GPU.
Numerical 16.2: Length of an AI-generated music track.
A track is set to 120 beats per minute in 4/4 time (4 beats per bar), with 32 bars.
To fit an exact 60-second video, you would ask for 30 bars (120 beats).
Numerical 16.3: Writing NPC lines at scale (illustrative).
A game has 200 NPCs, each needing 15 barks in 3 variations, which is 200 × 15 × 3 = 9,000 lines. Writing from scratch takes 2 minutes a line; editing an AI draft takes 0.5 minutes.
Writers save 225 hours, which they can spend on the main story and character voice.
Numerical 16.4: Aspect ratios for social formats.
An image generator is asked for 1080 pixels wide. What height is needed for each format?
| Format | Ratio (W:H) | Height for 1080 px width |
|---|---|---|
| Instagram square post | 1:1 | 1080 |
| Instagram portrait post | 4:5 | 1080 × 5/4 = 1350 |
| Reels, Shorts, Stories | 9:16 | 1080 × 16/9 = 1920 |
| YouTube thumbnail | 16:9 | 1080 × 9/16 = 607.5, so usually 1280 × 720 is used |
16.6 Industry use cases
- Ubisoft Ghostwriter (2023). Ubisoft announced Ghostwriter, an in-house tool that generates first drafts of NPC "barks" (short lines said in response to events). Scriptwriters choose and edit the drafts, so the tool removes repetitive work while writers keep creative control.
- AI-generated art in competition. In August 2022, a Midjourney-generated image submitted by Jason Allen, Théâtre D'opéra Spatial, won first prize in the digital arts category at the Colorado State Fair, sparking a public debate about AI and art.
- Brand campaigns. Coca-Cola's "Create Real Magic" (Chapter 12) used GPT-4 and DALL·E to let artists create images with brand assets.
- Licensed training data. Adobe states that Firefly is trained on licensed content such as Adobe Stock and public-domain material, positioning it for commercial users concerned about copyright.
16.7 Review questions
- Explain how a diffusion model turns a text prompt into an image.
- Write an image prompt for an e-commerce listing of a brass diya, naming subject, setting, style, composition, lighting and format.
- Evaluate the benefits and concerns of AI music tools for a small video production firm.
- Design an AI-assisted workflow for creating NPC dialogue that keeps writers in control.
- How many bars are needed for a 45-second track at 100 BPM in 4/4 time? (Answer: 75 beats, so 18.75 bars; use 19 bars and trim.)
- Discuss the copyright position on AI-generated images, using the Zarya of the Dawn decision.
Knowledge sources
- Ramesh, A. et al., Hierarchical Text-Conditional Image Generation with CLIP Latents (DALL·E 2), 2022.
- Rombach, R. et al., High-Resolution Image Synthesis with Latent Diffusion Models (Stable Diffusion), CVPR 2022.
- US Copyright Office, Zarya of the Dawn registration decision, 21 February 2023; Copyright and Artificial Intelligence reports.
- Video Games Chronicle, Ubisoft is developing an AI tool that aims to support scriptwriters, March 2023 (reporting Ubisoft's Ghostwriter announcement).
- Digital Music News, UMG, WMG and Sony sue AI music companies Suno and Udio, 24 June 2024.
- Adobe, Adobe Firefly.
- Ministry of Electronics and IT, Government of India, advisories to intermediaries on deepfakes, 2023.
Independent study notes for learning purposes. Not affiliated with any university or course provider. Facts and case-study figures are linked to their original sources. ChatGPT, Gemini, Copilot, Jasper, Synthesia, Runway, Lumen5 and other product names are trademarks of their respective owners.