If you want a video your class will actually watch, start with ScreenPal to record yourself explaining something on your own screen, and Edpuzzle to turn any video into something you can check. Those two cover the work most teachers mean when they say "make a lesson video." The AI video generators being marketed to teachers this fall, the ones that build a clip from a text prompt, sit at the far end of this list for a reason: the moment a person's face or voice is involved, you have a consent question before you have a production question.
This is for K-12 teachers making their own lesson media: a recorded explanation for students who missed class, a diagram for a handout, an audio version of a dense reading, a video with questions built into it. It is not about slide decks. If that is the job, the AI presentation tools post is the right page.
The four jobs hiding inside "make a lesson video"
Teachers search for one thing and need one of four, and the four want different tools.
Job one: record yourself. You talk through a worked problem, a text, a lab procedure, with your screen showing. Your voice, your slides, your annotations. This is the most common job and the least glamorous.
Job two: check that they watched. You have a video, yours or someone else's, and you want questions inside it and a record of who answered what.
Job three: make the picture. A labeled diagram, an image of a historical scene, a visual for a worksheet. No video at all, but the same request in a teacher's head.
Job four: generate the video. You describe a scene and a model builds it. This is the newest and the one that carries the most conditions.
Sort your task into one of those four before you compare anything. A tool that is excellent at job one is often useless at job three, and the vendor pages rarely say so.
How this list was built
The pool was the 11 tools in the directory's Presentations, Design and Media category, plus tools recorded under other categories whose actual job is media: Edpuzzle sits under Assessment and Grading, and Gemini Notebook under the directory's tutoring category. From that pool, seven tools are covered here because each one owns one of the four jobs above and has a published price or a published free tier.
The criteria: does it do one of the four jobs without a second tool bolted on, does the vendor publish what it costs, does the vendor publish something checkable about student data, and does the teacher keep the judgment. Nobody pays to be included on this blog and no vendor saw this list before it published.
What was left out, and why. SlidesAI and Prezi AI make slide decks, which is a different job with its own page. Figma for Education is a collaborative whiteboard rather than a media maker. Kleveroo builds review games. AI STL Generator makes 3D print files, which is a real classroom job and a narrow one. The harder exclusion: several general AI video generators now advertise to teachers, and a search for the best one returns page after page of near-identical sites wrapping the same underlying model. Only one of that family, Dance2AI, is in the directory with a published price, so that is the one treated here, and it is treated as the category rather than as a recommendation.
Pricing and privacy positions were checked against each vendor's own pages on September 24, 2026.
What the directory records show about media tools
Of the 138 tools listed on aieducator.tools, 11 sit in Presentations, Design and Media, and only 5 of those 11 state a data hosting location, a FERPA position, a COPPA position or a minimum age. That is the worst disclosure rate of any sizable category on the directory, where 93 of 138 tools state a FERPA position overall.
The pattern inside that gap is the interesting part. The six media tools with blank records include the four most widely used in classrooms: Canva for Education, Adobe Express, ScreenPal and ThingLink. The two that do claim FERPA and COPPA outright in our records, SlidesAI and KnowledgeFlow, are the two almost nobody has heard of. So the tools a district is most likely to already have deployed are the ones our own records say least about.
That gap is fixable from published sources, and this post fixes it for the seven tools below. All four of those vendors document a position on student data on their own sites. The directory records should carry it, because that is the first place a teacher checks.
The seven, compared
| Tool | Best for | Free tier | Paid from | Student data position | Hosting stated |
|---|---|---|---|---|---|
| ScreenPal | Recording your own screen | 15-min cap, 10 videos | $3.00/mo (Deluxe Edu) | iKeepSafe FERPA certified; school official | United States |
| Edpuzzle | Checking who watched | 20 saved activities | No readable public price | iKeepSafe FERPA, COPPA, CSPC seals | Not stated |
| Canva for Education | Visuals plus simple video | Free for verified K-12 | Free for eligible schools | States FERPA and COPPA certified | Not stated |
| Adobe Express | Generating images | 5GB, daily image generations | Free K-12 license (Premium) | Firefly toggle per district | Not stated |
| ThingLink | Making an image clickable | 30-day teacher trial | $9/mo billed annually | FERPA school official exception | Not stated |
| Gemini Notebook | Audio version of a reading | Yes, limits unpublished | Not published on the page checked | Not used for training unless you give feedback | Not stated |
| Dance2AI | Generating video from a prompt | No | $8/mo billed annually | None offered for schools | United States |
The student data column is what each vendor says about itself, not an audit. Every figure in it was read from the vendor's own page on September 24, 2026.
ScreenPal is the best fit for recording your own explanation, because its free tier hosts the video as well as making it
Most teachers do not need a video generated. They need fifteen minutes of themselves talking through a problem, hosted somewhere a student can reach it. ScreenPal records screen and webcam together, adds microphone narration, and gives you a link. The free plan hosts the file, which matters more than it sounds: a tool that records but leaves you to upload the result somewhere else adds a step you will not repeat weekly.
The AI features, auto-transcription, captions in over 100 languages, chapter generation and video summaries, sit on the paid education tiers. Captions are the part worth paying for. An uncaptioned lesson video is inaccessible to some of your students and unsearchable for all of them.
What sets it apart from Canva, the obvious alternative: Canva will record you, but its center of gravity is design, and long-form screen capture with quizzes is not what it is built around. ScreenPal is a recorder that grew editing, not a design suite that grew recording.
Watch out for: the free plan's 15-minute recording limit and 10 hosted videos. Fifteen minutes is a defensible ceiling on pedagogical grounds. Ten hosted videos is not, and you will hit it inside a semester.
Pricing: free plan $0/mo with a 15-minute recording limit, 10 hosted videos and 5 video quizzes. Deluxe Edu $3.00/mo paid annually for unlimited recording and hosting. Max Edu $6.00/mo paid annually, which is the tier that includes the AI features. Team Education $3.00/mo per user with a 10-user minimum (education plans, checked September 24, 2026).
Safety: ScreenPal's privacy policy, dated July 22, 2025, states that it "is hosted in the United States", that it "holds the iKeepSafe FERPA and CSPC Certification", and that when it collects information directly related to an identifiable K-12 student it processes that data "as a 'School Official' under FERPA" (privacy policy). The policy says ScreenPal does not direct its services to children under 13 or knowingly collect their personal information "unless and except when ScreenPal is used by a school", and that in the EU it is not intended for under-16s. The school-official route means your district, not you, is the party agreeing to that arrangement, so check whether your district has.
Edpuzzle is the best fit for knowing whether a video was watched, because the questions live inside the video rather than beside it
A video you assign and cannot check is a hope. Edpuzzle takes a video, yours or one from YouTube, and lets you cut it, drop questions at chosen timestamps, add your own voice narration over it, and see per-student answers and viewing data in a gradebook.
For the absent-student problem this is the half that ScreenPal does not solve. You record the explanation once, put three questions in it, and you have evidence the student engaged rather than left it playing in another tab. It is the same problem the AI tools for substitute teachers post solves for a planned absence, from the other end.
What sets it apart from ScreenPal's video quizzes: Edpuzzle's questions are the product rather than a feature, so the reporting is built for a teacher with five sections rather than for a creator checking engagement.
Watch out for: the free plan stores 20 activities total, not 20 per class. Teachers are offered three more slots per verified colleague referred, which is a growth mechanism dressed as a feature. Plan on hitting the ceiling.
Pricing: the free Basic plan lets you "save up to 20 activities in your account", raised by 3 per verified colleague who joins (help center, checked September 24, 2026). The Pro Teacher price is unverified: edpuzzle.com/pricing did not render a readable plan table when it was fetched on September 24, 2026, and the help center describes the tiers without naming a figure. School and district licensing is quote-based.
Safety: Edpuzzle's COPPA direct notice to educational institutions, dated July 22, 2024, states that it "collects the minimal amount of information from students necessary to create accounts in our service", that schools and districts "may act as the parent's agent and consent to the collection of students' personal information", and that Edpuzzle "does not sell or rent students' personal information ... to any third party for any purpose, including for advertising or marketing purposes" (COPPA notice). It holds the iKeepSafe FERPA, COPPA and CSPC seals. The notice does not address whether student data is used to train AI models, which is the question a 2026 procurement review will ask, so ask it.
Canva for Education is the best fit for the teacher who needs one tool for both the picture and the short video, because the whole premium product is free if you qualify
Canva earns its place on breadth. Its directory record carries a 4.62 rating from 39 reviews, the largest review count of any media tool listed, and for an eligible K-12 teacher the full product costs nothing: templates, 1TB of storage, the Magic Studio AI tools, and assignment handoff into Google Classroom, Canvas and Schoology.
For job three, making the picture, Canva is usually the fastest route from idea to something printable, and for a short video it will do the job without you learning an editor.
What sets it apart from Adobe Express, its closest rival: both are free for K-12 and both generate images. Canva is stronger on templates and classroom workflow; Adobe is stronger on the image model itself and on what it will tell you about that model's training data.
Watch out for: eligibility is narrower than teachers expect. Canva's published guidelines exclude homeschool educators unless they are licensed with a K-12 institution, along with public librarians, higher education teachers and parents of K-12 students, and require eligible users to "re-verify their credentials every 3 years" (eligibility guidelines). Build a department workflow on it and someone's access lapses in year three.
Pricing: free for certified K-12 teachers and staff, school librarians, learning support assistants and curriculum specialists, and free for accredited K-12 schools, districts and departments of education. Access lasts "for as long as you're an active and eligible teacher or organization" (eligibility guidelines, checked September 24, 2026).
Safety: Canva states that "Canva for Education is certified compliant with FERPA" and "certified compliant with COPPA", and that it "complies with COPPA by not knowingly collecting personal information from a child under 13 unless a school or district has authorized us to collect such information in order to provide educational services" (trust for education). That page does not state where student data is hosted, and it does not say whether student content is used to train Canva's AI models. Both are reasonable questions for a district to put in writing before a rollout.
Adobe Express is the best fit for generating images you will put in front of students, because Adobe publishes what the model was trained on
For job three, the question that decides it is not image quality. It is whether you can answer a parent asking where the picture came from. Adobe publishes that: Firefly is "trained on Adobe Stock images, openly licensed content, and public domain content, where the copyright has expired" (Firefly for K12). No other image generator in this pool states its training corpus that plainly.
Adobe also gives district administrators a switch. The same page states that Firefly generative AI features are "available and turned on by default for all K12 educators and students", with a toggle controlling access to all of them and a separate toggle for the text-generation features. Default-on is a choice a district should know it is inheriting.
What sets it apart from Canva: the published training corpus and the admin toggle. What Canva has instead is the classroom workflow.
Watch out for: the consumer free plan and the education license are very different products, and teachers land on the consumer one by accident. The consumer free tier gives 5GB of storage, free daily image generations, but only "5 free lifetime uses" of presentation generation and "2 free lifetime uses" of video generation (Express pricing). Two lifetime video generations is a demo, not a tier. Get the school license.
Pricing: free for K-12 schools and districts. Adobe describes "a free Adobe Express for K-12 Education license that gives you access to Adobe Express Premium at no cost" (Adobe Express for Education). Outside that route, Adobe Express Premium is US$9.99/mo with 250 generative credits per user per month, and Adobe Firefly Pro is US$19.99/mo (Express pricing, checked September 24, 2026).
Safety: the directory record is blank on FERPA, COPPA, hosting and minimum age. Adobe's K12 documentation covers the admin controls and the training corpus but does not state whether student prompts are used for training, so treat that as unanswered. Adobe describes "education-specific privacy controls" on its education page without detailing them there. The practical control you do have is the Firefly toggle, and it sits with your district administrator rather than with you.
ThingLink is the best fit for making one image do the work of a worksheet, because the interaction is added to the picture rather than around it
Some material is a single image that needs twelve labels: a cell, a map, a Renaissance painting, a piece of lab equipment. ThingLink lets you drop hotspots onto an image, a video or a 360-degree photo, each opening text, audio, another video or a 3D model, and embed the result in your learning management system.
This is the quiet one on the list, and the job it does has no good substitute. Building the same thing in Canva means a slide per label. In ThingLink it stays one picture the student explores.
What sets it apart from Canva or Padlet: the unit is a single enriched image rather than a page or a board, and it supports 360-degree media and augmented reality without code.
Watch out for: the useful features sit above the teacher license. AI-assisted course creation, automatic translation, 360-degree video, augmented reality authoring, single sign-on and LTI integration are all listed under the school Enhanced plan, which is quote-only (K-12 plans). A teacher paying personally gets the interactive images and not much of the AI.
Pricing: teacher license $9/month billed annually, after a 30-day free teacher trial. The three school and district plans, Essential, Enhanced and Staff Training, are custom priced with no published figure (K-12 plans, checked September 24, 2026).
Safety: ThingLink's privacy policy, dated June 1, 2026, states that it "is authorized by schools and districts under the FERPA 'school official' exception to receive and use educational data to provide educational services", that children "may only access the Services at the direction of their teacher or school district by using an invitation code", and that it does "not use personal data of students for the purposes of targeted marketing or advertisements" (privacy policy). It describes its databases as behind a firewall but does not state their geographic location, and it does not mention AI training. Two unanswered questions, both worth asking.
Gemini Notebook is the best fit for turning a reading into audio, because it will only work from the sources you upload
Job three has a cousin nobody plans for: a text some students cannot access as text. Gemini Notebook, which Google previously called NotebookLM, takes your PDFs, documents, slides, URLs and audio and produces study guides, quizzes, flashcards and an audio overview, all with inline citations back to your sources.
The audio overview is the feature that matters here. Upload the chapter your class is struggling with and you have a discussion-style audio version students can listen to on the bus, grounded in that chapter rather than in whatever the model knows. For multilingual classes, pair it with the translation tools in the AI tools for English language learners post.
Being source-bound is the pedagogical point. The model summarizes what you gave it, so a claim in the output can be traced to a page in your text. That is the opposite of asking a chatbot what it thinks about photosynthesis. Google has also been shipping teacher training on this family of tools monthly, adding modules on the first Wednesday of every month since at least September 3, 2026 (Google AI Educator Series).
What sets it apart from ChatGPT for the same task: the citations and the source boundary. ChatGPT will happily answer beyond your document.
Watch out for: the age gates. Google states that users must be "above the age of consent", that age verification is required, that under-18 users face "stricter content policies" and that "certain features may only be available to users over 18" (about Gemini Notebook). For a managed school account, "your organization's admin needs to turn on access." Plan this as a teacher-facing tool that produces something for students, not as a tool you hand to a class of twelve-year-olds.
Pricing: there is a free tier. Google's own support page describing the product does not publish the free tier's usage limits or any paid tier, pointing instead to a separate usage-limits article, so no figure is quoted here (about Gemini Notebook, checked September 24, 2026).
Safety: Google states that "your data is protected, and is not used to train Gemini Notebook unless you provide feedback", and for Workspace users that "your uploads, queries and the model's responses in Gemini Notebook will not be reviewed by human reviewers, and will not be used to train AI models" (about Gemini Notebook). That is the clearest training-data statement of the seven. The age gate, not the data handling, is what constrains classroom use.
Dance2AI is the category to understand rather than the tool to buy, because prompt-to-video is where the consent problem lives
Dance2AI generates video from a text prompt or a still image, with reference control and 500 to 4,800 credits a month depending on tier. It is competent at what it does and it is the wrong shape for a classroom, and it is here because the category it represents is being marketed hard to teachers this fall.
Two things to notice. First, searching for the best AI video generator for teachers returns a swarm of near-identical sites, most of them wrapping the same underlying model under different domain names, which makes the category almost impossible to evaluate by search. Dance2AI is one of the few with a directory record and a published price, which is why it stands in for the category here.
Second, and this is the real point: a prompt-to-video tool has no school data agreement, no student-facing mode, and no reason to have thought about either. Its directory record lists FERPA and COPPA as not applicable, a minimum age of 13, and United States hosting. Not applicable is accurate. It is also the answer that should stop a district rollout, because it means nobody on the vendor side has considered your obligations.
What to choose instead: for a short animated explainer, Canva or Adobe Express under your school license, with the school's data terms already in place. Use a general video generator for a thing you made about your subject, never for anything involving a student.
Watch out for: likeness. The moment you upload a photo of a person to an image-to-video tool, you are generating a moving image of that person on a consumer platform. If the person is a student, stop.
Pricing: Lite $8/month billed annually, regular price $10/month, with 500 credits, up to 50 videos and 250 images a month. Pro $24/month billed annually with 2,000 credits. Mega $48/month billed annually with 4,800 credits. Credit packs from $9 (pricing, checked September 24, 2026).
Safety: the directory record states United States hosting, a minimum age of 13, and FERPA and COPPA as not applicable. There is no education tier, no data processing agreement for schools and no student mode. Treat anything you make in it as published to a consumer service.
The consent test: three questions before you press record
This is the part the tool comparisons leave out, and it is the part that decides whether a lesson video is a good idea. Ask three questions in this order.
Who appears in it? If the answer is only you, your legal position is simple and your next decision is pedagogical. If a student appears, audibly or visibly, you are now handling something the US Department of Education may treat as an education record. Its guidance states that a photo or video of a student is an education record when it is "(1) directly related to a student; and (2) maintained by an educational agency or institution" (FAQs on photos and videos under FERPA). A video can be the education record of several students at once, and if a parent asks to inspect it, the school must redact the other students "without destroying the meaning of the record" where it reasonably can. A recording of your screen and your voice creates none of this. A recording of your third period does.
Whose face or voice is being synthesized? Generating a likeness is a different act from recording one, and districts have begun writing rules about it specifically. If a tool clones your voice or animates a face, get that in writing from your district before it reaches students, and never put a student's image into a generator.
Where does the file live? A video hosted by the vendor sits under that vendor's terms. ScreenPal says it hosts in the United States and acts as a school official. Canva, Adobe Express, ThingLink and Edpuzzle do not publish a hosting location. If the video contains a student, that unanswered question is the one that matters, and the FERPA-compliant AI tools post covers what to ask a vendor.
Run those three questions and most of the anxiety around AI video in schools resolves into a simple rule: make media about your subject, not media out of your students.
What about ChatGPT, MagicSchool, Diffit, Khanmigo and Quizlet?
These are the names that come up whenever teachers ask about AI media, so here is where each lands on the four jobs.
ChatGPT is the best of the general assistants for scripting a video, and its video generation is a consumer feature with the same likeness problem as any other. Use it to write what you will say, not to produce what students see.
MagicSchool is a strong teacher toolkit and not a media tool: it will write the script, the questions and the differentiated version, then hand you off. Its directory record carries a 4.74 rating from 19 reviews, and it earns that on text.
Diffit reworks a reading to different levels and handles translation, which solves the access problem that an audio overview also solves, by a different route. Reach for it when the barrier is reading level rather than modality.
Khanmigo is a tutor, not a production tool. It is the right answer to a student who is stuck, and no answer at all to a teacher who needs a diagram by Thursday.
Quizlet has no page on the directory. It is a study-set tool, and for the "check they watched it" job Edpuzzle is the closer fit because the checking is attached to the video.
What this looks like across one week
Here is a plan for a seventh-grade science unit on cell structure, using four of these tools in sequence, the kind of sequence the AI tools for middle school teachers post sets out more broadly. This is a plan, not a report of a class that ran.
Monday. Record one 12-minute ScreenPal video walking through the diagram you will use all week, with your annotations appearing as you talk. Twelve minutes fits inside the free tier's 15-minute cap. Name it by lesson, not by date, so it survives into next year.
Tuesday. Build the same diagram as a ThingLink image with a hotspot on each organelle, one line of text and one 20-second audio clip per hotspot. This becomes the reference students return to, and it replaces the labeling worksheet rather than supplementing it.
Wednesday. Push Monday's video through Edpuzzle with three questions: one at 4 minutes, one at 8, one at the end. Assign it to the students who were absent and to anyone who asked for it again. You now have per-student evidence rather than a raised hand.
Thursday. Upload the textbook chapter to Gemini Notebook and generate the audio overview. Offer it to students for whom the chapter is a wall. Listen to it first: the point of a source-bound tool is that you can check it against the source, so check it.
Friday. Make one image in Adobe Express under your school license for the assessment sheet, and nothing else. Resist the urge to generate a video.
One measure of whether the week worked: how many students who missed Monday could label the diagram on Friday.
How to choose, in three questions
Which of the four jobs is this? Record yourself, check they watched, make the picture, or generate the video. Job one is ScreenPal. Job two is Edpuzzle. Job three is Adobe Express for a generated image, Canva for a designed one, ThingLink for one that needs labels, Gemini Notebook when the barrier is reading rather than seeing. Job four is a category to approach carefully and usually to skip.
Does a student appear in it? If yes, you are working with something that may be an education record, and the vendor's hosting location and data terms stop being paperwork. If no, choose on fit and price.
Will you still be doing this in March? This is the question that kills the free tiers. Ten hosted videos and 20 saved activities are semester-long allowances at best. Work out the real annual cost before a department standardizes on a free plan, because the migration later costs more than the license now.
For UK and Australian readers, the same four jobs apply to cover lessons and remote learning, and the compliance frame is GDPR and your data protection officer rather than FERPA. The consent test survives the change of jurisdiction intact.
One last thing worth knowing before you make anything. Research on instructional video is unusually settled: Mayer's review sets out fourteen evidence-based design principles, including segmenting a complex visual into progressively presented parts, signaling the key material, presenting words as spoken rather than printed text, using conversational language, and, pointedly, not displaying a static image of the instructor's face while gesturing on camera works better (Mayer, 2021, Journal of Applied Research in Memory and Cognition, 10(2), 229-240, abstract). A twelve-minute screen recording where you talk and annotate satisfies most of that list. A generated video of a synthetic presenter satisfies almost none of it.
The bottom line
Record yourself with ScreenPal, make it checkable with Edpuzzle, make your pictures in Adobe Express or Canva under a school license, label them in ThingLink, and use Gemini Notebook when a student needs to hear a text rather than read it. Skip prompt-to-video for anything involving a person. The full directory of 138 tools is there if you want to compare the rest, and the six-of-eleven disclosure gap above is the reason to read the vendor's own privacy page before your district signs anything.
Dan Fitzpatrick is a Forbes contributor, three-time bestselling author and founder of The AI Educator, and has trained more than 150,000 educators across 30+ countries. More from Dan.
