Where to Get Graphics for eLearning: Stock, AI, and Custom
By Devlin Peck · Updated
Part of the eLearning Design & Development guide
You don't need to draw a single graphic to build great-looking eLearning. Instructional designers get their visuals one of three ways: they pull them from stock and illustration libraries, they generate them with AI, or they lightly customize what they find. Below you'll get the full source list with licensing notes, plus a framework for deciding when AI generation helps and when it hurts.
Graphics are worth sourcing carefully because they carry the moments that learners remember most. My favorite part of any project is adding moments of delight, using animation, interaction, and storytelling to create something memorable, and the visuals you pick set what's possible for those moments. Sourcing is one piece of a bigger process, so if you're building your skills end to end, start with my guide on how to design effective eLearning.
Where can you get graphics for eLearning projects?
The best sources are free illustration libraries (Freepik, Storyset, unDraw), free stock photo sites (Pexels), eLearning-specific character libraries (eLearningArt), paid stock photos (Envato Elements, Adobe Stock), and the AI generators built into tools like Storyline 360. Here's the full list at a glance:
| Source | Best for | Cost | License notes |
|---|---|---|---|
| Freepik (now Magnific) | Vectors, photos, and AI generation in one place | Free tier + paid plans | Free tier requires attribution; paid plans remove it |
| Storyset | Matching illustration families you can recolor in-browser | Free tier + paid plans | Free tier requires attribution (it follows Freepik's licensing) |
| unDraw | Open-style illustrations recolored to your palette | Free | Commercial use allowed, no attribution required |
| IllustrationKit | Clean, modern illustration sets | Free | Personal and commercial use, no attribution required |
| Black Illustrations | Illustrations of Black people for diverse representation | Free packs | Free for personal and commercial use |
| Pexels | Royalty-free photography | Free | Free for commercial use, no attribution required |
| eLearningArt | Photographed character cutouts for scenarios | Paid membership | Built specifically for eLearning use |
| getIllustrations | One-time illustration pack purchases | Free samples + paid packs | Commercial use with purchased packs; check each pack's terms |
| Envato Elements | High-volume, high-polish stock of every asset type | Paid subscription | Unlimited downloads with commercial license while subscribed |
| Adobe Stock | Premium photography and vectors | Paid credits or subscription | Standard license covers typical course use |
Sites, tiers, and license notes checked as of August 2026. License terms change, so confirm on the source's own license page before you ship client work.
One naming note so you don't get confused mid-search: Freepik rebranded as Magnific in 2026 as it shifted toward a full AI creative platform, but the library you know (and its free tier) is still there.
Should you use stock, AI-generated, or custom graphics?
Use stock when speed and polish matter most, use AI when you need something specific that stock doesn't have, and customize stock assets when brand consistency is non-negotiable. Here's how the three approaches compare:
| Approach | Speed | Cost | Style consistency | Licensing risk | Best for |
|---|---|---|---|---|---|
| Stock (free or paid) | Fast | Free to moderate | High within one library or family | Low if you follow attribution rules | Most course visuals, photography of real people |
| AI-generated | Fast to iterate, slow to perfect | Low | Hard to keep consistent across many images | Output isn't copyrightable; training data varies by tool | Concept scenes, backgrounds, styles stock doesn't have |
| Custom-edited stock | Slower | Your time | Highest, since you control colors and composition | Same license as the source asset | Brand-matched courses, flagship portfolio projects |
The licensing point often surprises people. According to the US Copyright Office's copyrightability report, summarized by the Copyright Alliance, fully AI-generated images lack human authorship and can't be copyrighted. This cuts both ways: you can't stop anyone else from using an image you generated, and your client should know the course visuals they paid for aren't protectable the way commissioned artwork is.
What are the best free graphics sites for eLearning?
If you only bookmark three, make them Magnific, Storyset, and unDraw. Together they cover vectors, matching illustration families, and recolorable one-off illustrations. Here's the full rundown:
Magnific (formerly Freepik)
Still the workhorse. Magnific has an enormous library of vectors, photos, PSDs, and now AI generation under their expanded brand. The catch on the free tier: you must credit the author, and per their own licensing docs, only paid subscribers can skip attribution. Download the vector formats when you can, because they're the ones you can pull apart and recolor.
Storyset
Storyset is Magnific's illustration platform, and it solves the hardest free-graphics problem: consistency. Each concept comes in themed families, and you can change colors, backgrounds, and elements in the browser before downloading, no Illustrator required. There's also a Figma plugin if that's where you design. It's my first answer to "how do I get illustrations that match across a whole course?"
unDraw
unDraw offers open-style illustrations that are free for commercial use with no attribution required, which makes it the lowest-friction option on this list. Set your brand color once and every illustration downloads already recolored.
IllustrationKit
IllustrationKit has clean, premium-feeling illustration sets that are free for personal and commercial projects with no attribution required. A solid alternative when the big libraries all start looking the same.
Black Illustrations
Black Illustrations provides free illustration packs of Black people for digital projects, free for commercial and personal use. If your learner population is diverse and your graphics aren't, this is an easy fix.
Pexels
Pexels is my go-to for free, royalty-free photography with no attribution required. Photography beats illustration when learners need to see real environments, real equipment, or real faces.
I walked through most of these sites on screen in this video. It's from 2021, but every site in it is still live as of this update:
What are the best paid graphics sources?
Paid stock images makes sense when you need volume, photography of real people, or eLearning-specific assets that free libraries don't carry. The main options are Envato Elements (an unlimited-download subscription across photos, vectors, video, and templates), Adobe Stock (premium assets via credits or subscription), getIllustrations (one-time illustration pack purchases, no subscription), and eLearningArt (the eLearning-native option, with 50,000+ character cutout images built for course scenarios).
I'll be upfront about my own stack, though: I canceled my Envato Elements and Adobe Stock subscriptions in 2026, because AI generation replaced them for our team. That's not a knock on either service. If you produce courses that need photography of real, consistent humans, or your client's legal team wants traditionally licensed assets, paid stock is still the right call. But it's no longer the default it was five years ago, which brings us to the real question.
When should you use AI-generated graphics, and when do they scream AI slop?
AI generation is now my team's primary graphics source, and it works best for concept scenes, backgrounds, and style-matched illustration sets you can't find in stock. It fails hardest on photorealistic people learners must trust across a course, brand-accurate assets, and anything technically precise: UI screenshots, diagrams, and safety procedures need to be captured or drawn, not hallucinated.
Here's what this looks like in practice for us. I don't do much image generation myself (I spend far more of my AI time on code than on pictures), but Sabrina on our team handles our visual design and uses Magnific for everything, running the newest image models it hosts, like Seedream and Nano Banana Pro. The workflow is iteration: generate, judge, adjust the prompt, regenerate, and only ship the version that would survive next to a stock asset without embarrassment.
Two more options worth knowing:
- In-tool generation. Storyline 360's AI Assistant generates images from text prompts with prebuilt styles right inside your project file, per Articulate's AI Assistant guide. It's useful beyond images, too. When I first tested it back in 2024, I asked it for futuristic, robotic font options for a sci-fi themed course title, and it pointed me to Orbitron, which was a legitimately good pick. It was a small decision, but real time saved.
- Commercially safe generation. For risk-averse clients who ask "where did this image come from?", Adobe Firefly's FAQ states its commercial models are trained on Adobe Stock, openly licensed content, and public-domain material. That provenance answer matters in regulated industries.
So how do you avoid the slop look? Check your output against these tells:
- Character drift. The same "person" changes hairstyle, age, or uniform between slides.
- Mangled text. Generated signage, labels, or UI text with garbled letters. Generate text-free images and add real text in your authoring tool.
- Uncanny details. Hands, teeth, and object physics still may betray AI output on close inspection.
- Style drift. One image photorealistic, the next flat illustration, the next a 3D render. This is the tell learners notice even when they can't name it.
How do you get consistent character images for scenario-based eLearning?
Use a photographed character library or a single illustration family, because a scenario falls apart the moment your protagonist looks different on slide 12 than she did on slide 3. eLearningArt's character library was built for exactly this: the same person photographed in dozens of poses and emotions, as cutouts you can drop onto any background. On the illustrated side, one Storyset family gives you a matching cast and world for free.
AI character generation is catching up but still takes work to hold consistency across images. OpenArt's character feature has been popular with the students I work with for building a consistent cast, though of the tools we've tried it has the steepest learning curve. Magnific's newer models handle character reference well too. Whatever you pick, generate your full cast with locked-down style prompts, then reuse that style throughout.
Consistent characters matter most in scenario-based learning experiences where the learner builds a relationship with the cast across many decisions. And once your cast looks consistent, the modern next step is letting learners actually talk to those characters. Full disclosure, devlin.ai is my company: you describe a scenario in plain English and get a working voice or text conversation simulation that embeds in Storyline, Rise, or any LMS.
How do you customize graphics to fit your course?
Download vector formats (SVG, AI, or EPS) whenever they're offered, because vectors recolor and rescale cleanly while raster images (PNG, JPEG) pixelate when you enlarge them. That single habit unlocks most customization. From there, the workflow is short:
Grab the vector version
On Magnific and similar sites, choose SVG, AI, or EPS over PNG when available. You can open these in Adobe Illustrator or Figma and select every individual element inside the graphic.
Recolor and recompose
Swap the graphic's colors for your course palette, delete elements you don't need, and pull out individual icons or characters from larger scenes. This is how one stock asset becomes ten on-brand assets.
Export as PNG for your authoring tool
Export with a transparent background as PNG, not JPEG, so your graphic sits cleanly on any slide background. Export at 2x the display size so it stays crisp.
If you're new to vector editing, this walkthrough covers the extract-recolor-export workflow from scratch:
You don't strictly need Illustrator. Figma handles SVG editing well, and even PowerPoint can ungroup and recolor many vector files. For a broader look at what belongs in your toolkit, see my roundup of instructional design software and authoring tools.
What rules apply no matter where your graphics come from?
Three rules outlast every tool on this page: each graphic must earn its place instructionally, each course gets one visual family, and every graphic must clear accessibility requirements.
Every graphic earns its place. Research in the multimedia learning tradition backs this up: Sung and Mayer's study on graphics in online lessons found that relevant graphics improved learning, while decorative ones improved how much learners liked the lesson but not what they learned. Source against the learning goal, not the vibe.
Irrelevant graphics make learners like your course more. They don't make them learn more.
One visual family per course. Decide on your illustration family or photo style before you build slide one, and hold every asset to it. This applies to small defaults, too: if you're using Storyline's built-in shapes for buttons, go with sharp rectangle corners instead of the rounded defaults, because sharp corners match the buttons on real modern websites and keep the whole project feeling current. You can see this discipline in every one of the standout eLearning examples I've collected: each commits to a single visual world.
Meet accessibility requirements. WCAG success criterion 1.4.11 requires essential non-text elements like icons and UI components to have at least a 3:1 contrast ratio against adjacent colors. And apply alt-text discipline: instructional graphics get meaningful alt text describing what the learner needs to take from them, while purely decorative graphics get empty alt text so screen readers skip them.
Sourcing well is half the job; arranging what you sourced is the other half, and that's covered in my guide to visual design principles for eLearning. And once the finished course is ready for learners, my guide on how to use an LMS covers getting it delivered and tracked.
Frequently asked questions
Do I need to give attribution for free graphics?
It depends on the site. Freepik and Storyset require attribution on their free tiers, while unDraw, IllustrationKit, Pexels, and Black Illustrations allow commercial use with no attribution. Always check the source's license page before shipping, and keep a record of what the terms said when you downloaded.
Who owns AI-generated images?
Under the US Copyright Office's current position, fully AI-generated images lack human authorship and aren't copyrightable, so nobody owns them in the copyright sense. You can generally use them in your courses (check your tool's terms), but you can't stop others from using similar output, and clients should know this before you build their course on AI visuals.
Can I use graphics from these sites in client work?
Yes, in most cases. Every source in the table above permits commercial use on at least one tier, but watch two things: free tiers that require attribution (awkward in client deliverables) and restrictions on using assets in logos or trademarks. When in doubt, a paid tier or license usually removes the friction.
What file format should I download for eLearning graphics?
Choose vector formats (SVG, AI, EPS) whenever they're available, since you can recolor and resize them without quality loss. For import into Storyline or Rise, export your final graphic as a PNG with a transparent background. Use JPEG only for photographs that fill their whole frame.