Multimodal AI — Creating Beyond Text

Module 7 · Images, Video, Audio, and More

Content current as of April 2026. AI tools evolve rapidly — underlying principles remain applicable.

Learning Objectives

Builds on: Module 4 (prompting), Module 5 (verification)

After completing this module, you will be able to:

Explain how AI generates images, video, and audio — in plain language
Write effective prompts for image generation tools
Identify the strengths and limitations of each multimodal modality
Combine text, image, audio, and video AI tools into a creative workflow
Recognize AI-generated content and understand deepfake risks
Apply ethical guidelines to multimodal AI creation

The Pope in a Puffer Jacket

In March 2023, an image went viral showing Pope Francis wearing a stylish white Balenciaga puffer jacket. It was shared millions of times. News outlets scrambled to confirm or deny it. For a few hours, a significant portion of the internet believed the Pope had suddenly become a fashion icon.

The image was entirely AI-generated. A Reddit user named Pablo Xavier had created it using Midjourney in about 30 seconds. When asked how he made it, he said he was “just messing around.”

The Pope image marked a turning point: the moment the general public realized that AI could create images indistinguishable from photographs. Not blurry, not obviously fake, not “AI-looking.” Photorealistic images that fooled millions of people, created in seconds by someone with zero design training.

This module covers the entire multimodal landscape — images, video, audio, and more. The capabilities are extraordinary. So are the responsibilities.

Multimodal AI in YOUR Field

AI-generated anatomical diagrams for patient education
Medical imaging AI that highlights areas of concern on X-rays and CT scans
Text-to-speech for accessibility in patient communications
Video walkthroughs of post-surgical care instructions

Product photography generated from text descriptions (no photoshoot needed)
Brand identity packages: logos, color palettes, marketing visuals
Video ads and social media content at scale
Presentation visuals generated from slide content

Illustrations for course materials generated from learning objectives
Audio summaries of readings for students with visual impairments
Educational animations explaining complex processes
Multilingual content creation for diverse classrooms

Drone imagery + AI for crop disease identification
Infographics generated from yield data for stakeholder reports
Training videos for equipment operation and safety procedures
Visual documentation of field conditions for compliance

Common Misconceptions About Multimodal AI

“AI images are always detectable” — Detection is getting harder. The Pope jacket fooled millions. AI detection tools have significant false positive and false negative rates.
“AI art isn’t creative” — The AI isn’t creative. The PERSON directing it can be. Creativity lives in the vision and judgment, not the execution tool.
“Video AI is ready for production” — It’s ready for short clips, b-roll, and concepts. Full production-quality video still needs human direction and editing.
“AI-generated voice is always fake-sounding” — Modern text-to-speech is nearly indistinguishable from human voice. Voice cloning from a few seconds of audio is real and raises serious ethical concerns.
“If I generated it, I own it” — Copyright for AI-generated content is unsettled law. Adobe Firefly offers the clearest commercial licensing. For other tools, check the terms.

How AI Image Generation Works

The Short Version

Image generation models (DALL·E, Midjourney, Stable Diffusion, Ideogram, Flux) use a technique called diffusion. Here’s the intuition:

Start with pure noise — random colored pixels, like TV static
Gradually remove the noise — step by step, guided by your text prompt
Each step makes the image slightly more coherent — shapes emerge, colors form, details sharpen
After hundreds of steps — a fully formed image appears

It’s like a sculptor starting with a rough block of marble and chipping away until a figure emerges — except the sculptor is guided by your words.

The model learned what images look like by training on billions of image-text pairs from the internet. It knows that “sunset over mountains” typically involves orange/pink skies, silhouetted peaks, and warm light — because it’s seen millions of examples. When you prompt it, it’s not searching a database. It’s generating a new image that matches the patterns associated with your words.

Image Generation Tools

Tool	Strengths	Best For	Access
DALL·E 3 (OpenAI)	Text rendering, integration with ChatGPT	Quick generation in conversation	ChatGPT Plus
Midjourney	Artistic quality, distinctive aesthetic	Professional creative work	Discord or web
Stable Diffusion	Open-source, local running, customizable	Maximum control, privacy	Free, runs locally
Ideogram	Best text-in-image rendering	Logos, signs, text-heavy designs	Web, free tier
Flux	High quality, fast	Quick high-quality generation	Various platforms
Adobe Firefly	Commercial licensing clarity	Business/commercial use	Adobe Creative Cloud

Writing Prompts for Images

Image prompting is different from text prompting. The key variables:

Subject: What’s in the image (“a golden retriever puppy sitting in autumn leaves”) Style: The artistic approach (“watercolor painting,” “photorealistic,” “flat vector illustration,” “anime style”) Composition: How it’s framed (“close-up portrait,” “wide landscape,” “bird’s eye view”) Lighting: The mood (“soft golden hour light,” “dramatic studio lighting,” “neon cyberpunk glow”) Details: Specific elements (“wearing a red collar,” “with a mountain background,” “in the style of Wes Anderson films”)

Example — from vague to specific:

❌ Vague: “a dog”

✅ Better: “a golden retriever puppy sitting in a pile of autumn leaves, soft afternoon sunlight filtering through trees, shallow depth of field, warm color palette, photorealistic style”

Guided Practice 7.1

Using any image generation tool you have access to (DALL·E through ChatGPT, Ideogram, or Bing Image Creator — all have free options), generate images for:

A professional headshot-style portrait of a fictional character for a business website
An infographic-style illustration explaining how photosynthesis works
A logo concept for “Roadrunner Lemonade” (the brand from Module 6)

For each, write at least two prompt versions — one simple, one detailed. Compare the results.

What to notice

The detailed prompts produce dramatically more usable results. Also notice what the AI struggles with: text in images (often garbled), hands (sometimes wrong number of fingers), and consistent character design across multiple images. These are known limitations of current models that are improving rapidly.

AI Video Generation

Current State (April 2026)

Video generation has made remarkable progress but remains less mature than image generation:

Tool	What It Does	Quality Level
Sora (OpenAI)	Text-to-video, up to 60 seconds	High quality, sometimes uncanny
Runway Gen-3	Text-to-video + video editing	Good for short clips
Kling (China)	Text-to-video, strong at motion	Competitive quality
Veo (Google)	Text-to-video, integrated with YouTube	Improving rapidly

Realistic Use Cases Today

Short social media clips (5-15 seconds)
B-roll footage for presentations and videos
Concept visualization (“show me what this product would look like in use”)
Educational animations (simple demonstrations of concepts)
Storyboarding (generating visual frames for a planned video)

What Video AI Can’t Do Yet

Consistent characters across scenes (the person looks different in each clip)
Complex physics (water, hair, and fabric often look wrong)
Long-form narrative (anything over 30-60 seconds breaks coherence)
Reliable text or UI elements in video

✅ Test Yourself: Multimodal Basics

How do diffusion models generate images? (Start with noise, gradually remove it guided by the text prompt)
Name one thing AI image generators still struggle with. (Hands/fingers, text in images, consistency across multiple images)
What’s the current practical limit of AI video? (Short clips 5-60 seconds; consistent characters and complex physics are still weak)

AI Audio and Music

Text-to-Speech

Modern AI can generate speech that is nearly indistinguishable from human voices:

Natural narration for presentations, videos, and podcasts
Voice cloning — replicating a specific person’s voice from a few seconds of audio sample
Multilingual dubbing — translating AND speaking in another language while matching the original speaker’s voice

Music Generation

Tools like Suno and Udio generate original music from text descriptions:

“Upbeat acoustic folk song with guitar and harmonica, about a road trip through California, male vocals, tempo 120 BPM”

The output is a complete song — vocals, instruments, production — generated in about 30 seconds. Quality ranges from “surprisingly good” to “obviously AI” depending on the genre and complexity.

NotebookLM Audio Overviews

As covered in Module 5, NotebookLM can generate podcast-style audio discussions of your uploaded documents. Two AI voices have a natural-sounding conversation about your material — useful for reviewing content while commuting or exercising.

Combining Modalities: The Creative Pipeline

The real power of multimodal AI emerges when you chain tools together:

Example Pipeline: Course Marketing Video

1. TEXT → Write a script using ChatGPT/Claude
2. TEXT → IMAGE → Generate visuals for each scene with DALL·E/Midjourney
3. TEXT → AUDIO → Generate narration with text-to-speech
4. IMAGE + AUDIO → VIDEO → Combine in a video editor (or use Runway)
5. REVIEW → Human review for quality, accuracy, and appropriateness

Example Pipeline: Brand Identity Package

1. TEXT → Brand Sprint prompt (from Module 6) → positioning, tagline, colors
2. TEXT → IMAGE → Logo concepts using image generation
3. TEXT → DOCUMENT → Brand guidelines document using AI writing
4. TEXT → IMAGE → Social media templates using brand colors/fonts
5. REVIEW → Human refinement and final selection

The person with this workflow isn’t a designer, videographer, copywriter, or musician. They’re a director — someone who knows what they want and can guide AI tools to produce it. The bottleneck is no longer skill. It’s taste, judgment, and intent.

Recognizing AI-Generated Content

As AI-generated content becomes more sophisticated, recognizing it becomes an essential literacy skill:

Visual Tells (as of 2026)

Hands and fingers — still occasionally wrong (extra fingers, merged digits)
Text in images — often garbled or misspelled
Background inconsistencies — objects that blend or defy physics
Texture repetition — identical patterns in hair, fabric, or foliage
“Too perfect” quality — an uncanny smoothness that photos don’t have
Symmetry that’s too exact — real faces and objects are slightly asymmetrical

Detection Tools

Tools like Content Credentials (Adobe), AI-or-not, and Hive Moderation attempt to detect AI-generated images and text. They’re useful but imperfect — false positives (marking human content as AI) and false negatives (missing AI content) are common. Don’t rely on any single detector.

The Deepfake Concern

AI-generated video and audio of real people (deepfakes) pose serious risks:

Political manipulation — fake videos of candidates saying things they never said
Financial fraud — fake CEO video calls authorizing wire transfers (this has actually happened)
Personal harassment — fake explicit content of real people
Erosion of trust — when anything could be fake, nothing feels trustworthy

The Ethics of Multimodal AI

Creating AI-generated images, video, or audio of real people without their consent is unethical and often illegal. Creating misleading content designed to deceive is always wrong, regardless of the technology used. The Pope puffer jacket was humorous. A deepfake of a political candidate is dangerous. The technology is the same — the intent and impact are what matter.

Module 7 Summary

Key Takeaways

Image generation uses diffusion — starting from noise and gradually shaping it guided by your prompt. Detailed prompts with subject, style, composition, and lighting produce far better results.
Video generation is advancing rapidly but still limited to short clips. Best for b-roll, concepts, and social media — not long-form narrative.
Audio AI can generate near-human speech, clone voices, and create original music. Powerful for content creation, risky for deception.
The creative pipeline chains modalities: text → image → audio → video. The bottleneck is no longer skill — it’s taste, judgment, and intent.
AI-generated content can often be recognized by visual tells (hands, text, texture), but detection is imperfect and getting harder.
Deepfakes pose real risks: political, financial, and personal. Creating deceptive content of real people is unethical and often illegal. (See Module 10 — Ethics for a deeper analysis of AI ethics and fairness.)
Your creative judgment — knowing what’s good, appropriate, and honest — is now the most valuable skill in the room.

💭 Think About It

Multimodal AI means anyone can create professional-quality content — images, video, music, entire brand identities — with zero traditional training. This is democratizing in the best sense: a small business owner in any city can create marketing materials that rival a New York agency’s output. But it also means the barrier between real and fake has collapsed. A world where anyone can generate a photorealistic image of anything means verification isn’t just an AI skill — it’s a survival skill. Every image, every video, every audio clip you encounter now requires the question: Is this real?

🔬 Lab 7: The Brand Sprint

Lab 7: The Brand Sprint (Full Version)

Time: 60 minutes | Tools needed: Any AI chatbot + any image generation tool (free options: DALL·E via ChatGPT, Bing Image Creator, Ideogram)

Purpose: Design a complete brand identity from scratch using AI — the full BA 1028 workflow.

Part A — The Brief (15 min) 1. Choose a real or hypothetical project: a campus club, a side business, a nonprofit event, or a personal brand. 2. Use the Brand Sprint prompt from Module 6. Let the AI interview you with 8 questions about your brand. 3. Review the AI’s brand output: positioning statement, tagline variants, color palette, font pairing, and logo concept prompts.

Part B — Visual Identity (20 min) 4. Take the 3 logo concept prompts the AI generated and run them through an image generation tool. 5. Generate at least 2 variants of each concept (6+ images total). 6. Select your favorite 2-3 outputs.

Part C — Brand Sheet (15 min) 7. Ask the AI to create a one-page brand sheet: logo placement guidelines, color swatches with hex codes, font hierarchy, and tone of voice summary. 8. Compile everything into a single document or presentation.

Part D — Reflection (10 min) 9. Write 3-4 sentences: How long would this have taken without AI? What required YOUR judgment that AI couldn’t handle? What would a professional designer do differently?

Submit: Your brand brief (AI interview), 6+ generated images, your brand sheet, and reflection.

🎮 Module 7 Quiz

Previous: Module 06 — AI for Productivity | Next: Module 08 — Agents, GPTs, and Automation