Explainer and product demo videos for software clients: what AI can do and what it must not touch

Software clients are a good fit for video and a risky fit for careless AI. The product is a screen. Every label, number and click on that screen is a promise to a buyer. If your video shows a button that does not exist, the client pays for it in support tickets and you pay for it in trust. So the question for an agency is not "can AI make this video". It is "which layers of the video can AI make without changing the truth".

This page is for the owner or head of a small agency. I use Higgsfield as the example for the AI scene layer, and I say plainly where other tools are the better pick. Prices are as of November 2026, so check each vendor's page before you quote a client.

This page may contain affiliate links from which Blur Studio earns commissions.

Explainer, demo or tutorial: pick the job first

Clients say "explainer" for three different videos. Agree which one you are making before you write a quote, because they cost different amounts and need different footage.

Video	Job	Typical length	Needs real screen recording?
Explainer	Say what the product is and why it matters, for someone who has never heard of it	60 to 90 seconds	A few seconds, if any
Product demo	Show the product doing one job from start to finish	90 to 150 seconds	Yes, most of the video
Tutorial	Teach a user to complete a task	2 to 6 minutes	Yes, every step

A rule of thumb for scripts: about 130 to 150 spoken words per minute. A 90 second video is roughly 200 to 225 words. Read the script aloud with a timer before you pay for any voice or clip. Sources disagree on the exact pace (one vendor says 130, another 150), so time your own read.

If the client wants a broader series of short answer videos rather than one flagship video, my page on an explainer video series covers that. For process explainers drawn as simple figures, see stickman explainer videos.

The two layers: the truth layer and the presentation layer

This is the rule I would put in your agency's playbook. Split every software video into two layers.

• Truth layer. Real screen capture of the real product with sample data, real clicks, real menus and real pricing text. Nothing here is generated.
• Presentation layer. Everything that frames the truth: the opening scene, abstract visuals, B-roll, motion on a still image, voice, captions, music, transitions.

AI belongs in the presentation layer. Writers who cover AI product demos make the same point: generated interfaces tend to invent labels, metrics and states that the product does not have. Do not count on a future model fixing that: you cannot check a generated screen against the product as easily as you can record the real one. Your duty is to the client's product, so the truth layer stays recorded.

Where AI helps and where it must not touch the UI

Part of the video	AI is a good fit	Use a human or a recording
Opening hook (a person stuck in a messy workday)	Yes, as a short generated scene	Filming or stock if the client wants real people
Abstract "data flowing" visuals	Yes	A motion designer if the brand look is strict
B-roll between steps	Yes	Stock footage is often cheaper for generic shots
Voiceover	Yes, with the licence checked	A voice actor for a flagship brand video
Captions and translations	Yes, with a native reviewer	A translator for legal or security text
The product screen, clicks and settings	No	Always recorded
Pricing, plan names, security or compliance claims	No	Client-approved text only
Customer quotes and logos	No	Real, with written permission

If the product concept is abstract (a developer tool with no visual output, for instance) an animation studio may do better than any AI pipeline, because the metaphor is the product. Say so to the client. A recommendation that can go the other way is worth more.

Script to storyboard to clips, step by step

1. Write the script with a claims column. Two columns: what the voice says, and the proof for it (a screen in the product, a page in the docs, or a named approver). No proof, no sentence.
2. Time it. Read it aloud. Cut to fit the length you sold.
3. Storyboard each line as "recorded" or "generated". Recorded lines get a click script (the exact steps and sample data). Generated lines get a one-sentence scene description. My page on storyboard to video shows the frame-by-frame method.
4. Record the truth layer first. Use a dedicated demo account with fictional data, no notifications and no secrets on screen. Removing secrets at the source beats blurring them later.
5. Generate the presentation clips. Make 3 to 5 attempts per scene and keep one. Plan for rejected attempts in your hours and in your credits (credits are the points an AI tool charges per generation).
6. Record or generate the voice last, after the client has signed off the claims. Re-voicing a changed sentence costs time and credits.
7. Edit, caption, export, and keep the raw takes and click script in the client folder for the refresh job later.

Prompt template for a generated hook scene, to adapt: "A [role] at a desk at the end of a long day, surrounded by sticky notes and open browser tabs, calm and slightly tired, soft daylight, no text on any screen, camera slowly pushes in, 5 seconds." The line "no text on any screen" matters: generated text is where tools make things up.

Voice and licensing

A synthetic voice is part of what you deliver, so the licence is part of your risk. ElevenLabs, an AI voice platform for text to speech, lists a commercial licence from its Starter plan ($6 a month after a $1 first month on 4 November 2026) and does not list one on the Free plan. Other voice tools differ, and free tiers of many tools do not include business use. Confirm each tool's current terms, keep a copy of the terms version on the project file, and ask the client whether they want the voice disclosed as synthetic. My page on using AI video commercially covers the general rules, and keeping the same voice across demo videos covers consistency.

If the client wants a talking presenter, Synthesia, a tool that makes videos of AI avatars reading a script, is built for that and publishes plans on its pricing page (on 4 November 2026 a plain page load showed Starter at $9 for the first month then $19 a month, or $14 a month billed $168 yearly, while a browser check earlier showed $29 and $18 a month billed $216 yearly; the figures differ between loads, so confirm them live). That is a better fit than assembling a presenter from generated clips. A voiceover over the product screen is the simpler default for a demo, so use an avatar only when the client asks for a face.

What studios charge, and what an AI-assisted job costs you

Be careful with price benchmarks, and treat all of these as information, not a quote. The ones you can find online come from sellers, and some are old.

• A 2022 Wyzowl survey of 116 quotes for a 60 second animated video, cited by the screen-recording vendor Vibrantsnap, gave a median of $5,400 and a range from $600 to $250,000. That survey is about four years old, and the vendor itself says to treat it as an order of magnitude.
• One animation studio estimates a software explainer at $2,400 to $3,600 over 7 to 10 weeks with two to three revision rounds.
• One AI video service lists AI explainers at $497 to $1,500 for 60 seconds, against $2,500 to $7,500 for custom 2D animation, and adds 25 to 40 percent for each extra 30 seconds.

Treat these as vendor ranges, not market facts. Claims such as "AI is 50 to 80 percent cheaper and 5 to 10 times faster" come from sellers with no method behind them, so do not repeat them to a client. Your client will compare deliverables, not just price: ask what is included (script, voice, rounds, captions, refresh) before you compare. If the client asks whether they can just use AI themselves, see scripts for that conversation.

My own advice on the cost side: what AI saves you is mostly the presentation layer. Scripting, recording the truth layer, editing and reviews still take a person. Expect savings on the clip and B-roll hours, not on the whole job.

Where Higgsfield fits for this work

Higgsfield is a web platform that puts many AI image and video models in one place. For a software explainer it is useful for three jobs: the opening scene, abstract and B-roll clips, and animating a brand illustration or still frame into a short clip. Its Cinema Studio feature, a tool for directing camera, lens and lighting, runs on Seedance 2.5, a video model, and, per Higgsfield's help centre, can make clips up to 30 seconds with built-in sound at up to 1080p. See my page on Higgsfield Cinema Studio for how that tool works.

It is not a screen recorder or a demo tool, and it should not be used to produce the product interface. Treat it as the clip maker for the presentation layer only.

Credit math for planning. Higgsfield publishes a per-clip credit cost for some models. As of November 2026 its pricing page lists Kling 3.0 (a video model from the company Kuaishou) at about 7 credits for a 5 second 720p clip and Seedance 2.0 (the earlier ByteDance video model) at about 22 credits for 5 seconds at 720p. Cinema Studio 4.0 cost is not published (it shows on the Generate button). Plans at the last check: Plus at $59 a month ($47 billed annually) with 1,200 credits, Ultra at $129 ($99 annually) with 3,000. Check the Higgsfield pricing page summary and the live page.

A made-up example. Your 90 second explainer has 8 generated scenes of 5 seconds. You allow 5 attempts per scene on a model costing about 7 credits for 5 seconds. That is 8 x 5 x 7 = 280 credits, roughly 23 percent of a Plus month. At the Plus price that is about $14 of credits for the whole generated layer. The real cost is your hours reviewing attempts, not the credits. My page on spending Higgsfield credits wisely shows how to measure cost per approved clip.

Two cautions. First, Higgsfield's terms say content you put in or make may be used to train its models, except for customers on an Enterprise agreement (its top, custom-priced tier). The terms give users no IP indemnity either, so you carry the rights risk. So do not upload a client's unreleased screens or confidential material to a standard Team or Scale plan without checking the client's NDA (non-disclosure agreement). Second, the free plan adds a watermark to every output, and the pricing table lists commercial use as not included for Free while the help centre says the terms do not restrict commercial use to paid plans. For client work, use a paid plan and ask support if the point matters to your contract. You can try a single paid month on Higgsfield for one pilot video before you change how the whole agency works.

Pricing it to the client and keeping your margin

Sell the video as a fixed package with a defined number of revision rounds, not as hours. Here is a made-up example, using round numbers so you can swap in your own.

Item (illustrative)	Hours
Script and claims table	3
Storyboard	2
Screen recording and edit	5
AI clips, with retries	3
Voice	1
Two revision rounds	3
Total	17

At a loaded cost of $60 an hour that is $1,020 of time, plus about $100 of tools and credits, so $1,120. Sold at $2,500 (a made-up price), the margin is $1,380, about 55 percent.

Now the risk. Allow five revision rounds instead of two. At 1.5 hours a round, that adds 4.5 hours, so 21.5 hours, $1,290 of time plus $100 of tools, $1,390 total. Margin drops to $1,110, about 44 percent. The fix is in the contract: state the rounds included, define a round as one consolidated set of comments, and price extra rounds up front. See agency retainer pricing for the pricing models, and track the real hours so the next quote is better.

Review: the claims table and the accuracy pass

Before any voice is generated, send the client the script as a table.

Line	Claim	Proof in product or docs	Approver	Approved (date)
4	"Syncs with your accounting tool in minutes"	Docs page, integrations	Product lead
9	"Bank-level security"	Needs client legal review	Client legal

Security, compliance, uptime and legal claims go to the client's legal reviewer first. AI tools are a poor fit for regulated product claims and customer quotes. Then run an accuracy pass on the cut: every label on screen matches the live product, no fictional UI appears, no generated text is readable, and the sample data contains no real customer. Ask the client to confirm in writing that the demo environment holds no customer data. Pair this with the creative approval workflow so sign-off is recorded.

Keeping videos current after a redesign

Software changes, and a video that shows the old interface starts to cost the client. Turn this into a service. Keep the demo account, click script and raw takes in the client folder, then offer a refresh: re-record the changed screens, replace the affected lines of voice, and re-export. A suggested price is around 15 percent of the original project per refresh (my suggestion, not an industry figure). Because the presentation layer is generated clips and a licensed voice, only the truth layer and a few voice lines change.

Who should skip this approach

• Agencies whose client wants real customers on camera. Film them.
• Products in regulated fields where every frame needs legal review. The review time wipes out the AI saving.
• One-off videos where a screen recorder with built-in captions and a voiceover is enough. A tool like Vibrantsnap, a screen-recording product, may be all you need, and no AI clips at all.
• Agencies that cannot spare the hours to check generated clips. AI clips only save time if someone reviews them quickly.

How I'd approach this

I would run one pilot on one willing software client. Record the real product first. Add a 5 second generated opening and a few B-roll cutaways. Use a licensed voice. Send the claims table before voicing. Track hours and credits, then decide whether to sell it as a standard package.

Once it works twice, the work is repeatable: the claims table, the click script, the refresh offer and a shared folder of takes. Blur Studio builds the repeatable workflow, the client approval portal and the shared asset library around tools like these, if you would rather not assemble it yourself. See how I work with agencies.