Before we get into AI image generation — a quick gut check on where the images around you actually come from.
Not all AI image generators work the same way. The same prompt produces very different results depending on where you run it — and knowing those differences is a professional skill.
Each platform is free with a standard account. Some have daily generation limits — log in before class starts so your team doesn't hit a wall mid-activity.
Trained on licensed Adobe Stock images — outputs are commercially safe. Best in class for clean professional graphics and logo marks.
Uses OpenAI's image generation model. Creates images from prompts or uploaded images. Strong at following instructions, rendering text, and refining images through conversation.
Powered by Google's Imagen models. Accessible with any Google account. Excellent at understanding natural language prompts and creating vibrant, highly detailed artistic styles.
Google's workspace-oriented generation engine. Ideal for rapid prototyping and iteration. Offers clean composition presets that make it great for layout and design planning.
Powered by DALL-E. Sign in with any Microsoft or Outlook account. Strong at photorealistic scenes and a solid starting point for beginners.
The same prompt produces noticeably different outputs across tools. Here's what to watch for when you compare your results:
| Platform | Strengths | Best For | Notes |
|---|---|---|---|
| Adobe Firefly | Professional graphics, illustrations, and clean vector-style artwork. | Branding, logos, marketing graphics, and design assets. | Trained on licensed Adobe content for commercial-friendly workflows. |
| ChatGPT Images | Excellent prompt understanding, image editing, and conversational refinement. | Photorealistic scenes, illustrations, concept art, and iterative design. | Can improve images through multiple prompt revisions. |
| Gemini | Strong natural language understanding with colorful, detailed results. | Creative concepts, illustrations, and artistic imagery. | Works well from simple, conversational prompts. |
| Google Flow | Visual storytelling and rapid creative exploration. | Storyboards, scene planning, and video concept development. | Designed more for visual storytelling than standalone graphics. |
| Microsoft Copilot | Fast, photorealistic image generation with simple prompts. | General-purpose image creation and realistic scenes. | Excellent starting point for beginners. |
In today's lab you'll write prompts for two completely different image types. Understanding what each requires shapes how you write your prompts.
Used for storyboarding advertisements, website hero images, and TV commercial planning. The goal is a convincing, specific scene — the right mood, lighting, setting, and subjects. Vague prompts produce generic stock-photo results. Specificity is everything.
Used for brand identity work. The goal is a clean symbolic image that could represent your company. AI struggles with text in logos, so the strategy is to generate the visual mark only — no wordmark — and evaluate the symbol itself. Your company name gets added separately by a human designer.
Writing a prompt for an image generator is completely different from writing a text prompt. You're not describing what you want to say — you're describing what you want to see.
Professional image prompts are built in layers. Think of it like briefing a photographer — the more specific you are, the closer the result is to your vision.
Start with the main subject — the person, object, scene, or concept that should dominate the image. Be specific: not "a woman" but "a woman in her 30s in a business suit." Not "a city" but "a downtown skyline at dusk."
Weak: a coffee shop
Strong: a small independent coffee shop interior with exposed brick walls, warm Edison bulb lighting, and a single barista behind a wooden counter
Specify the visual style. Without a style instruction, AI defaults to a generic stock-photo look. Naming a specific aesthetic gives the model a clear target.
Examples: photorealistic · cinematic film still · flat vector illustration ·
editorial photography ·
minimalist logo mark
For logo prompts, flat vector or minimalist icon consistently
produces cleaner, more usable results than photorealistic styling.
Lighting is one of the most powerful mood-setters in photography and advertising. AI responds well to specific lighting language. Mood words help the model select color temperature, shadows, and composition.
Lighting: golden hour sunlight · overcast soft light · dramatic side lighting · neon-lit night scene · clean studio white
Mood: energetic and bold · calm and aspirational · gritty and authentic · clean and minimal
Composition tells the AI where to place things in the frame. Use photography and filmmaking terms — the models understand them well.
Shot types: wide establishing shot · close-up portrait · overhead flat lay · worm's-eye view · rule of thirds
For logos: specify centered on white background
or isolated symbol, no background
to get a clean, usable mark.
Professional prompts include technical specifications — aspect ratio and exclusions. The "negative prompt" (telling the AI what NOT to include) is one of the most powerful tools available, though not all free platforms support it explicitly.
Aspect ratio: 16:9 landscape for website banner
· 1:1 square for social ·
9:16 vertical for mobile
What to exclude: For logos, always include no text, no words, no letters
— AI-generated text in images is almost always garbled and unusable.
The AI has almost nothing to work with. You'll get something generic — probably a stock-looking crowd scene with no specific mood, framing, or brand feel. Every team's result would look identical.
Strong — All five layers presentNow the AI has a subject, style, mood, lighting, composition, and technical constraints. The output will be specific enough to use as a real storyboard reference.
AI will almost certainly generate garbled text and a cluttered graphic. The name will likely be misspelled or distorted. Never ask AI to render your company name as text inside an image.
Strong — Symbol only, clear styleBy describing what the symbol should represent rather than what it should say, you give the AI room to generate a usable visual mark. Your team evaluates the symbol — the wordmark (the name) gets added separately by a human designer.
Your team will write two prompts, test each one across four AI platforms, and build a comparison slide deck — the kind of deliverable a creative agency would bring to a client meeting.
Use the builders below — one for your photorealistic scene, one for your logo mark. Generate, read aloud as a team, and refine before moving to Phase 2.
↑ Choose a prompt type to generate it.
Divide the platforms among your team members. Use the exact same prompt on every platform so you can make a fair comparison. Save or screenshot every result—even the unsuccessful ones. You'll use these images in your comparison presentation.
Firefly_Photo_v1.png). You'll need
them all for the deck.
Your deck is the deliverable — what you'd present to a client to explain which AI tool to use and why. Use Google Slides, PowerPoint, or Canva. Follow this required structure:
| Criteria | 5 — Excellent | 3 — Adequate | 1 — Poor |
|---|---|---|---|
| Prompt Accuracy | Image matches every element of the prompt | Most elements present, some missing or altered | Image bears little resemblance to the prompt |
| Visual Quality | Professional, usable in a client presentation | Acceptable but would need editing before client use | Unusable — artifacts, distortion, or garbled output |
| Client Usability | Could be used as-is in a storyboard or pitch | Useful as a direction reference, not a final asset | Not useful even as a reference |
One photorealistic scene prompt, one logo mark prompt — both refined by the whole team before testing.
Both prompts × four platforms. Every result saved and labeled in your shared Drive folder — good and bad alike.
A 6-slide presentation with side-by-side results, platform ratings, and a written team recommendation for a client.
AI image generation is one of the most contested areas in creative media right now. Professional photographers, illustrators, and designers are asking real questions — and the industry doesn't have settled answers.
Stock photographers and illustrators have already seen significant income loss as AI-generated images replace work that used to require a human. A 30-second prompt can produce an image that would have cost a client hundreds of dollars in licensing or thousands in a photo shoot.
Every new technology displaces some workers while creating new roles. Photography replaced portrait painters. Desktop publishing replaced typesetters. AI requires human prompt writers, creative directors, and curators — new jobs that didn't exist before.
AI models were trained on millions of images created by human artists — often without their consent and without compensation. The tool that's replacing their income was built on their work.
Discussion: If you were running a small communications agency, would you use AI image tools for client work? What would guide your decision?
For a small business or student-run agency, AI image tools dramatically lower the cost of visual production. A storyboard that used to require a hired illustrator can now be roughed out in an hour. A logo exploration that used to cost thousands in agency time can be prototyped for free.
Smaller businesses that couldn't afford professional creative work can now access usable imagery. This democratizes visual communication — organizations with no creative budget can still produce professional-looking materials.
When clients realize they can get "good enough" for free, the market for professional creative work contracts. Rates drop. Entry-level creative jobs — the ones you're currently training for — are the first to disappear.
Discussion: As a communications student entering this field, what's your honest reaction to the cost savings argument? Does "good enough" concern you personally?
There's currently no legal requirement in the US to disclose that an image was AI-generated when used in advertising or brand materials. But some argue that professional ethics should require it — especially when the client believes they're paying for original creative work.
Your agency uses AI-generated images to rough out a TV commercial concept for a client presentation. The images are clearly placeholders — they'll never go public.
An AI-generated image goes live in a client's social media campaign. The client doesn't know it's AI-generated. The agency billed for "original photography."
Your agency uses AI to generate 20 logo concepts in an afternoon and presents them as the work of your design team. The client pays a design fee. No AI disclosure is made.
You use AI-generated images in a portfolio piece you're presenting as your creative work. A future employer sees it and assumes you produced it yourself.
Discussion: Which of these feels clearly acceptable? Which feels clearly wrong? Where's the line — and who draws it?
Key terms from today's session — AI image generation, prompt engineering, and visual production vocabulary.
AI systems trained on large datasets of images that can produce new images from text descriptions. These models learn statistical patterns from millions of human-made images and recombine those patterns in response to a prompt. They do not "draw" or "design" the way a human does — they generate statistically likely visual compositions based on their training data.
A written description submitted to an AI image generator describing the desired visual output. Effective image prompts differ from text prompts — they describe visual elements (subject, style, lighting, composition, mood) rather than asking questions or requesting information. Prompt specificity is the primary determinant of output quality.
The underlying technology behind most modern AI image generators, including DALL-E, Stable Diffusion, and Adobe Firefly. Diffusion models work by learning to gradually remove "noise" from random pixel patterns to produce coherent images — essentially running image degradation in reverse. The result is a model that can generate highly detailed, realistic images from text descriptions.
An instruction telling the AI image generator what to exclude from the output. On platforms that support negative prompts, you can specify elements to avoid — "no text," "no watermarks," "no distorted hands." Not all free platforms expose this control directly, but including exclusion language in the main prompt achieves a similar effect on most models.
A style instruction telling the AI to produce an image that resembles a photograph rather than an illustration or graphic. Specifying "photorealistic" in a prompt significantly shifts the output toward convincing, camera-like imagery with realistic lighting, textures, and depth. Used in advertising and storyboarding when the goal is to approximate how a real scene would look.
A sequence of images used to plan a visual production — a TV commercial, a photo shoot, a film scene, or a website layout. Storyboards communicate visual direction before expensive production begins. AI-generated images are increasingly used as storyboard references, letting teams rapidly prototype visual concepts without the cost of actual photography or illustration.
A logo mark (or "brand mark") is the symbol or icon component of a logo — the Nike swoosh, the Apple apple. A wordmark is the text component — the company name in a specific typeface. Most professional logos combine both. AI can produce compelling logo marks but cannot reliably render readable text, which is why logo prompts should always specify "no text" and focus on generating the visual symbol only.
A visual style instruction for logo and icon prompts. Flat vector design uses simple geometric shapes, solid colors, and minimal detail — no gradients, shadows, or photorealistic textures. Preferred for logos because it scales cleanly at any size, works in single colors for embroidery or engraving, and avoids the distortion artifacts common in complex AI-generated imagery.
The proportional relationship between an image's width and height (e.g., 16:9, 1:1, 9:16). For professional AI image generation, specifying the aspect ratio ensures the output matches the intended use: 16:9 for website banners and TV storyboards, 1:1 for social media posts, 9:16 for vertical mobile content. Including aspect ratio in your prompt tells the model how to compose the scene within those proportions.
The dataset of images an AI model learns from during development. Most image generation models were trained on billions of images scraped from the internet — including work by professional photographers, illustrators, and designers, often without their knowledge or consent. Whether this constitutes copyright infringement or unfair use of creative labor is actively being litigated as of 2025.
The legal terms under which AI-generated images can be used for paid client work, advertising, or public distribution. Licensing terms vary by platform: Adobe Firefly is trained on licensed content and explicitly allows commercial use; other platforms have more ambiguous terms. Before using an AI-generated image in professional client work, always review the platform's current commercial use terms — these are changing rapidly as legal cases develop.
The skill of writing precise, effective instructions for AI systems to produce desired outputs. In image generation, prompt engineering involves layering subject, style, mood, composition, and technical constraints into a single prompt that consistently produces professional-quality results. As AI tools become standard in creative workflows, prompt engineering is increasingly regarded as a billable professional skill — the ability to brief an AI tool effectively has real commercial value.