Two overlapping workflows: using AI to generate original images, and using design tools to compose, layout, and communicate visually. They work best together — AI creates raw material, design tools turn it into something intentional.
Prompting AI to create original images from scratch. Strong for concept art, mood reference, project visuals, and custom imagery where stock photos don't exist. Output quality depends almost entirely on prompt quality.
Assembling images, type, and layout into a finished visual — a poster, a presentation graphic, a project handout, an exhibition label. AI generates assets; design tools turn them into communication.
Most projects use both: generate a background image with AI, bring it into Canva, add typography and project info, export a finished graphic. The seam between tracks is where the interesting creative decisions happen.
Collect 3–5 images that gesture toward what you want — not to copy them, but to identify what they have in common. Is it the color temperature? The amount of negative space? The texture? Those observations become your prompt vocabulary.
Lead with mood and emotional register before you describe what's in the image. "Melancholic and still, as if something just left the room — a bare wooden chair near a window" is stronger than starting with the chair. The emotional frame shapes every other decision the model makes.
Don't re-roll the same prompt hoping for a different result. Study what came back: what's wrong, specifically? Too bright? Wrong medium? Off mood? Change one element of the prompt, re-run, compare. Prompt iteration is design iteration.
A short paragraph you keep open while prompting. Anchors your session and prevents drift across multiple generations.
Generates images directly in the chat window. Strongest for concept-level visuals and when you want to iterate quickly with natural-language feedback. Also excellent at writing detailed prompts for use in other tools.
Generate · Prompt-write · IterateHighest aesthetic quality of any image generator for artistic and conceptual work. Access via Discord or midjourney.com. Output is a 4-image grid — use U1–U4 to upscale the one you want. Strong style vocabulary support.
High-quality concept artTrained on licensed content — commercially safer than most generators. Strong for realistic and photographic styles. Integrates with Photoshop and Express. Adds Content Credentials so viewers can see an image was AI-generated.
Licensed-safe · PhotorealisticThe fastest path from a generated image to a finished graphic. Drag in your AI-generated image, add text, choose a layout, export. Built-in AI generation, background remover, and Magic Edit. Strong for posters, slides, and handouts.
Compose · Layout · ExportAdobe's free-tier design tool. Cleaner typography control than Canva and tighter integration with Firefly-generated images. Good for exhibition panels, presentation graphics, and print-ready PDFs. Included with most school Adobe licenses.
Compose · Print-ready outputMore powerful than Canva for systematic design — use it when you need a consistent visual system across many assets (exhibition labels, slide templates, consistent poster series). Steeper learning curve but much better control over grids, spacing, and type.
Visual systems · Multi-assetCards with a purple top border are graphic design tools — composition rather than generation.
What is this image for? What should a viewer feel? Which Norman level is primary — visceral (immediate sensory hit), behavioral (how they interact with it), or reflective (what it means after)? Write 4–6 sentences. This brief goes into every AI session for this visual.
Search for images that gesture toward your emotional target — not to copy, but to identify what specific visual properties you're chasing. Save them in a folder. Before your first generation, describe what they share: "All of them have very little color. The light comes from one side only. There's a lot of empty space."
Those observations are your first promptSubject → Medium → Style → Mood → Technical → Exclusions. Start with the mood layer and work outward. If you're unsure how to describe a style, describe a reference image to Claude and ask: "What's the visual style of this? Give me 3–5 terms I could use in an image generation prompt."
Run the first generation. Compare against your brief, not against a vague feeling. Write down what's wrong specifically: "Too bright — the mood I want is heavier." "The medium looks digital, not painterly." "Good composition but the color is pulling happy when I need melancholic." One issue at a time.
Change one thing in your prompt and regenerate. Keep a running log of what you changed and what it did. After 3–5 rounds, you'll have a much clearer picture of what variables actually matter for your specific vision. Save the prompt that got closest.
Before placing an AI-generated image in a final project, know: Who owns it? Can you use it in this context? Different tools have different policies — Firefly images are generally safer for broad use than Midjourney. Document your prompt and which tool generated it. In an academic or exhibition context, you should always be able to explain how an image was made.
The generated image is raw material. Import it into your design tool, crop to the right aspect ratio, set typography, add any other elements. The AI generates the asset; the design tool turns it into communication. Don't skip this step — a strong image in a weak layout still reads as unfinished.
Ask Claude: "What typography would complement this image's visual tone?""Make it aesthetic" or "something cool and dark" produces generic results. AI image models respond to specific visual vocabulary — medium, style, mood, lighting — not to vague approval-seeking language.
Generating the same prompt 10 times hoping a better version appears is random search, not design. You'll get variation in the wrong dimensions while the core problem stays fixed.
Almost no AI-generated image goes straight into a final project. It needs to be cropped, color-graded, composed with type, sized for its context. Dropping a raw generation onto a slide or poster shows — the edges don't match the intent.
In a design or engineering course — and especially in a public exhibition — AI image use without acknowledgment reads as evasive. Reviewers and faculty will notice and will ask. Not having an answer is worse than the conversation itself.
Canva's built-in AI generation is convenient but lower quality than dedicated generators. Using Midjourney to create a poster (when it can only generate an image) means wrestling with text and layout in a tool that doesn't support it.