🖼️ Lesson 6.2: Magic Media — Text-to-Image & Text-to-Video

Last lesson, the intern drafted your layout and your words. Today we hand the intern a camera and a paintbrush. Magic Media generates images, short video clips, and graphics from a text description — you type what you imagine, and Canva paints it. Used well, it gets you a custom picture that no stock library has. Used carelessly, it produces six-fingered people, gibberish signs, and the exact same glossy "AI look" everyone else is posting. This lesson is mostly about the craft of the prompt and the honesty about the limits — because the difference between a usable AI image and an embarrassing one is almost entirely in how you ask and how critically you look. Same theme as always: AI is your intern, not your boss.

📚 What You'll Learn

By the end of this lesson, you will be able to:

Generate images from text with a reliable prompt recipe: subject + style + detail + mood
Choose styles and aspect ratios, and iterate a prompt toward a usable result
Generate short videos and graphics from text, and understand their current limits
Recognize where AI images fail — text-in-image, hands, brand accuracy, consistency
Apply ethics and attribution sense, and know when not to use AI images at all

⏱️ Estimated Time: 55 minutes

🎯 Project: Write and iterate 2–3 prompts to generate one genuinely usable custom image for your project, then place and edit it in a design.

In This Lesson

What Magic Media Is (and Isn't)

Magic Media is Canva's text-to-media generator. You'll find it in the left-side panel under Apps (or the ✨ Magic tools) as "Magic Media," with tabs for Image, Video, and often Graphic. You type a description, pick a style, pick a shape, click generate, and a few seconds later you get options. Click one and it drops straight onto your canvas as an ordinary element — you can resize it, crop it, filter it, put it in a frame, layer text over it, all the skills you learned in Module 4. An AI image is not a special locked object; once it's on your page it behaves like any other photo.

Here's the honest framing. Magic Media is brilliant for a specific job: getting a custom, imaginative image that doesn't exist in any stock library — "a watercolor fox reading a book in a cozy library," "an isometric illustration of a coffee shop," "a dreamy abstract background in teal and coral." It is not a replacement for a real photo of your real product, your real team, or a real event. If you need to show the actual croissant you actually sell, take a photo. If you need a whimsical background, a concept illustration, or a mood image, Magic Media shines.

🧠 The intern analogy, sharpened

Text-to-image is like an intern who is an astonishingly fast illustrator but has never seen the real world clearly. They'll paint you something gorgeous in thirty seconds — and give the person in it seven fingers, or write "COFFE" on the shop sign, and never notice. Their speed is real. So are their blind spots. Your job is to describe clearly, generate a few options, and look critically before you use anything. Never post an AI image you haven't inspected up close.

The Prompt Recipe — Subject + Style + Detail + Mood

The single biggest predictor of a good AI image is a good prompt. Vague prompts get vague, generic images. Specific prompts get images you can actually use. Here's a recipe that works almost every time — four ingredients, in plain language:

graph LR A["SUBJECT
what is it"] --> P["A clear,
usable prompt"] B["STYLE
how is it drawn"] --> P C["DETAIL
specifics & setting"] --> P D["MOOD
feeling, light, color"] --> P

Subject — the main thing: "a red fox", "a coffee cup", "a mountain lake".
Style — how it's rendered: "watercolor", "flat vector illustration", "3D render", "photorealistic", "pencil sketch", "isometric".
Detail — specifics and setting: "wearing tiny glasses, sitting in a library, surrounded by stacked books".
Mood — feeling, lighting, palette: "cozy, warm golden light, soft muted colors".

Put them together and you get something like: "A red fox wearing tiny round glasses, reading a book in a cozy library surrounded by stacked books, watercolor illustration, warm golden light, soft muted colors." That prompt gives Canva enough to make something specific and charming instead of generic.

Good vs bad prompts

❌ Weak (vague → generic)	✅ Strong (specific → usable)
"a coffee shop"	"cozy corner coffee shop interior, warm morning light through big windows, flat vector illustration, muted earth tones"
"a dog"	"a fluffy golden retriever puppy sitting in a field of daisies, photorealistic, soft natural light, shallow depth of field"
"a background"	"abstract flowing gradient background, teal and coral, smooth and calm, minimal, lots of empty space for text"
"a healthy meal"	"overhead flat-lay of a colorful grain bowl with avocado and roasted vegetables on a rustic wooden table, bright natural light, food photography style"

✅ Why this matters

Learning to prompt well is a genuinely transferable skill — the same clear-description muscle works in every AI image tool, not just Canva. And it's the cheap skill with the big payoff: better words in, better pictures out, with no upgrade required. Notice that the strong prompts read like you're briefing a human illustrator. That's exactly the right instinct.

💡 Tip — describe, don't command the text

Tell the model what should be in the scene, not what to write. Asking for a poster "that says GRAND OPENING SALE" almost always produces mangled letters (more on that below). Instead, generate a clean image with space for text, then add your real text in a Canva text box on top — crisp, correct, and editable.

Styles, Aspect Ratio & Iterating

Below the prompt box, Magic Media gives you two dials that change results a lot.

Style

Canva offers preset styles (Photo, Watercolor, Flat, 3D, Neon, Retro Wave, and more). Picking a style is often easier and more reliable than describing it in words — choose "Watercolor" from the buttons and you can drop that word from your prompt. Match the style to the job: flat vector for clean modern graphics, photo for realism, watercolor or illustration for warmth and whimsy. A mismatched style is the fastest way to make a design feel off.

Aspect ratio

Choose the shape before you generate — square for an Instagram post, portrait/vertical for a Story or Reel cover, landscape for a presentation or banner. Generating at the right ratio beats generating a square and cropping it later, because cropping throws away part of the image (and often the best part).

Iterating — the real skill

Your first generation is rarely the keeper. Treat it as a conversation, not a slot machine. Change one thing at a time so you learn what each word does:

Not warm enough? Add "warm golden hour light."
Too busy? Add "minimal, simple background, lots of empty space."
Wrong feel? Swap the style word: "watercolor" → "flat vector."
Close but not quite? Generate again — each click gives fresh variations of the same prompt.

⚠️ Watch Out — credits are limited

Every generation spends a credit. On the free plan you get a modest number of Magic Media generations per month; Canva Pro gives a much larger monthly allowance. Canva has adjusted these numbers over time, so treat the exact figure as "check it in your account," and treat the principle as fixed: generations are finite, so think before you click. Write the best prompt you can, pick the best style and ratio, then generate — rather than machine-gunning ten half-formed ideas and burning your month's credits by Tuesday.

Text-to-Video & Graphics

Text-to-Video

Magic Media can also generate short video clips from a text prompt — a few seconds of, say, "waves gently rolling onto a beach at sunset" or "steam rising from a coffee cup." These are handy as atmospheric backgrounds or B-roll to layer under text in the video editor (Lesson 5.2). But set expectations honestly:

Clips are short — a handful of seconds, not a finished film.
Motion can be strange — objects morph, physics wobbles, faces and hands misbehave even more than in still images.
It's credit-hungry — video generations typically cost more of your allowance than images, so use them deliberately.
It's best for abstract or atmospheric shots (nature, textures, moods), not for depicting specific people doing specific things.

The winning move is usually to generate a short clip as a background layer, slow or loop it, and put your real message on top in text — not to expect a narrative video out of one sentence.

Graphics

The Graphic option generates illustration-style elements — icons, stickers, spot art with clean edges — rather than photographic images. These are handy when Canva's (huge) built-in element library doesn't have exactly the little illustration you pictured. As always, inspect them: AI graphics can come out slightly lopsided or inconsistent, and Canva's existing library of real vector elements (Lesson 3.2) is often cleaner and free of credits, so check there first.

💡 Tip — reach for the free library first

Before you spend a credit generating an icon or a background, search Canva's built-in Elements and photos. There are millions of them, many free, and they cost you nothing. Save Magic Media for the times you need something that genuinely isn't in the library — that's where it earns its keep.

The Honest Limits & the Ethics

This is the most important section in the lesson. AI images are seductive because the good ones look effortless. But they fail in predictable ways, and using them badly can embarrass you — or worse, mislead people. Know the failure modes, and know when to walk away.

Where AI images reliably struggle

Weak spot	What goes wrong	What to do
Text in the image	Signs, labels, logos come out as garbled fake letters ("OPNEING")	Generate without text; add real text in a Canva text box on top
Hands & fingers	Extra fingers, fused hands, impossible poses	Avoid close-up hands; zoom past them; regenerate; or choose another image
Brand accuracy	Your exact logo, product, or packaging won't be reproduced correctly	Use a real photo of the real thing for anything brand-specific
Consistency	The "same" character looks different in every generation	Don't rely on AI for a recurring mascot across many images
Fine detail & counting	Wrong number of objects, warped small details, odd reflections	Inspect at full size before using; regenerate or fix in editing

⚠️ When NOT to use AI images — this part matters

Do not use AI-generated images to depict real events, real people, or facts. Never make a fake photo of a real person, a real news event, a disaster, a celebrity, or anything a viewer would reasonably read as "this really happened." That's how misinformation spreads, and it can seriously harm real people and your own credibility. Don't pass off an AI image as a genuine photo of your product or your team when the reality matters to the viewer. AI images are for imaginative, illustrative, and decorative work — concepts, moods, backgrounds, whimsy — not for standing in for the truth.

Ethics & attribution — the short, sane version

Be transparent when it counts. If your audience would care whether an image is real or generated (journalism, testimonials, "before/after" claims), say it's AI or don't use it. For a decorative background, nobody needs a disclaimer.
Respect real people. Don't generate identifiable real individuals, especially in situations they weren't in. It's disrespectful and often legally risky.
Know the usage rules. Canva sets terms for AI-generated content, including commercial use and restrictions; check Canva's current AI usage policy before you use generated media in paid work. The rules evolve — read them rather than assume.
Credit humans where their work is the point. AI doesn't replace paying a photographer or illustrator when you need genuine, authored, brand-defining work — it complements them.

🌟 Go further — the "would this fool someone?" test

Before you publish any AI image, ask one question: "Could a reasonable person mistake this for real, in a way that would matter?" If yes, don't use it (or label it clearly). If it's obviously an illustration, a mood, or a background — post away. That single question resolves the vast majority of AI-image ethics without needing a law degree.

🎯 Project: Prompt Your Way to a Usable Image

You're going to earn a real image the professional way — through iteration, not luck. Pick a design from your wishlist that could use a custom illustration or background (a social post, a flyer header, a presentation title slide). The goal is one usable image you actually made better across 2–3 tries, placed and edited in a design. Works on the free plan (mind your credits).

🏋️ From prompt to placed image

Objective: Generate a custom image with a deliberate prompt, iterate it toward better, place it in a design, and add your real text on top.

Instructions (about 35 minutes):

(5 min) Write your first prompt using the recipe: subject + style + detail + mood. Pick a style preset and the correct aspect ratio for your design before generating.
(5 min) Generate. Inspect the options at full size — check hands, edges, any text, and overall feel. Note honestly what's wrong.
(6 min) Iterate: change one thing (add light, simplify, swap style) and generate again. Do a third round if needed. Stop when you have one you'd be proud to post.
(4 min) Click your favorite to drop it onto the canvas. Resize, crop, or frame it to fit your layout (Module 4 skills).
(6 min) Add your real text in a Canva text box on top — not baked into the image. Make it readable with good contrast.
(5 min) Ethics check: is this clearly illustrative/decorative, not a fake "real" photo? Would anyone be misled? Fix or replace if so.
(4 min) Save the design, and save the winning prompt for reuse.

💡 Hint — a prompt worksheet

PROMPT RECIPE
Subject: ____________________
Style:   ____________________  (watercolor / flat vector / photo / 3D…)
Detail:  ____________________  (setting, specifics)
Mood:    ____________________  (light, color, feeling)

Aspect ratio: ____  (square / portrait / landscape)

ROUND 1 result — what's wrong? ____________________
Change ONE thing → ROUND 2 ____________________
Change ONE thing → ROUND 3 ____________________

INSPECT before using:
[ ] Hands / fingers OK?
[ ] Any text in image is garbled? (add real text on top instead)
[ ] Edges / small details clean?
[ ] Ethics: illustrative, not a fake "real" photo?

If three rounds don't get you there, that's useful data — it may be a job for a real photo or a stock image instead. Knowing when to stop is part of the craft.

✅ Project Completion Checklist

You wrote a recipe-based prompt and set style + aspect ratio before generating
You iterated at least twice, changing one thing at a time
You inspected the winner up close (hands, edges, any text)
You placed it in a design and added your real text on top, not baked in
You ran the ethics check and saved both the design and the winning prompt

🎯 Quick Quiz

Question 1: You need a poster image that says "GRAND OPENING." What's the smart move?

Question 2: Which is an appropriate use of an AI-generated image?

Best Practices

✅ Do's

Use the recipe. Subject + style + detail + mood turns a vague idea into a usable image.
Set style and aspect ratio first. Generating at the right shape beats cropping later.
Iterate one change at a time. That's how you learn what each word does — and get better fast.
Inspect before you use. Full-size check for hands, edges, garbled text, weird details.
Add real text on top. Keep words crisp and editable in a Canva text box, never baked into the image.

❌ Don'ts

Don't depict real events or people as if real. That's misinformation, full stop.
Don't fake your actual product or team. When reality matters to the viewer, use a real photo.
Don't ask the image to spell. Text-in-image is a known failure — do the words yourself.
Don't burn credits on lazy prompts. Think, then generate. Check the free library first.

💡 Pro Tips

Keep a "prompts that worked" file. Great images come from great prompts, and great prompts are reusable.
Generate abstract backgrounds with "lots of empty space" — they're the easiest AI win and give you room for text.
Combine tools: generate a mood background here, then use Magic Edit (Lesson 6.3) to clean up any oddities.
When you need brand-perfect accuracy, stop generating and take a photo. Knowing the tool's edge is expertise.

📓 Learning Journal

After this lesson, add to your journal:

Key concepts you learned
Techniques that clicked for you
Questions or confusion points to revisit
Ideas you want to try
Your progress and feelings about generating media with AI

✍️ This lesson's prompt: Write down your best-performing prompt and what you changed between rounds to get there. Then note one thing the AI got obviously wrong (hands? text? a weird detail?) and how you caught it. Finally, name one situation in your own work where you would deliberately choose NOT to use an AI image — and why. That last one is the professional's instinct, and it's worth putting into words.

📝 Lesson Summary

🎓 Key Takeaways

Magic Media generates images, short videos, and graphics from text, dropping them onto your canvas as ordinary, editable elements.
The prompt recipe is subject + style + detail + mood; specific in, usable out.
Set style and aspect ratio first, then iterate one change at a time; every generation costs a credit (Free = limited, Pro = larger allowance).
AI images reliably fail at text-in-image, hands, brand accuracy, and consistency — inspect before using, and add real text yourself.
Never depict real events or people as if real; AI media is for imaginative and decorative work, not for standing in for the truth. Check Canva's AI usage terms.

🎉 What You've Accomplished

You didn't just click "generate" — you learned to direct the tool: a repeatable prompt recipe, an iteration habit, an eye for the telltale AI mistakes, and an ethical line you won't cross. That combination is exactly what separates useful AI imagery from the generic, occasionally-embarrassing flood. You can now conjure a custom picture that no stock library holds — and you know when to put the tool down and reach for a real camera instead.

❓ Common Questions at This Stage

Can I use Canva's AI images commercially — in ads or products I sell?

Often yes, but with conditions, and the rules change. Canva publishes an AI product terms/usage policy that covers commercial use and prohibited content. Before you put generated media into paid or public work, read Canva's current terms rather than assume — and never use AI to depict real people or events, regardless of the plan.

My generated people look slightly "off." Am I doing something wrong?

Not necessarily — realistic humans (especially hands, teeth, and eyes) are the hardest thing for AI to get right, and small weirdness is common. Options: zoom past the tricky parts, regenerate, switch to an illustration style where imperfections read as artistic, or use a real stock photo of a person instead. For anything showing real people, a real photo is almost always the better call.

I ran out of credits. What now?

Credits reset on a cycle, so you can wait, or lean on Canva's free built-in photos and elements (millions of them) in the meantime. If you generate constantly for work, Pro's larger allowance may be worth it — but you can finish this course on Free by prompting thoughtfully and spending credits only when the library truly can't deliver.

🔭 Looking Ahead

Next — Lesson 6.3: Magic Edit, Grab, Eraser, Expand & Magic Switch — we stop generating from scratch and start editing existing photos with AI: brushing in changes, moving a subject off its background, erasing distractions, outpainting a photo to a new shape, and switching a whole design into other formats and languages. Several of these are Pro; we'll be crystal clear about which.

✅ Before the Next Lesson

Finish the project — one iterated, inspected, ethically-clear image placed in a design
Save your winning prompt to your prompts file
Have a photo handy (a product shot, a snapshot, anything) — we'll edit real photos next lesson
Write your Learning Journal entry, including your "when NOT to use AI" note

📚 Additional Resources

🌟 Encouragement for the Journey

You can now paint with words — and, just as importantly, you know when the painting should be a photograph. That judgment is the whole game. The internet is filling up with careless AI images; yours will stand out precisely because you prompt with craft, inspect with a critical eye, and refuse to mislead. Keep the intern sketching, keep your standards high, and let's go edit some real photos next. 🖼️