HRIS demos: screen + captions, not a twin HRBP

Overlay a talking founder you have, or stay faceless. An HRIS demo that has to ship is the product on screen — the workflow, the field, the approval — with words a muted viewer can read. It is not a generated HR business partner who never sat in your office.

Record the product. Caption it. If someone talks, it is someone you actually recorded.

The demo is the product

Buyers of HR software are screening the UI, not a culture film. They want to see the request, the manager step, the audit trail. A photoreal "HRBP" explaining the module is a person. If you invented them, you have a replica you did not hire. If you twinned a real employee without a replica grant, you have a likeness file, not a demo.

Stay inside the product. A screen recording is the base. Add captions for the beat the mouse cannot say. Timed text for the one line that has to sit on a field: "one click from request to approval," not a generated mouth restating it.

HR and talent teams already have the adjacent employer-brand problem: candidates discount a polished culture clip that could belong to anyone. The same discount applies to a demo. Specificity is the UI. A twin presenter is the table-tennis table.

A founder overlay is not a replica HRBP

If you want a face, use a face you have. Film the founder, the PM, or the CS lead against a plain backdrop. Picture-in-picture composites that clip onto the screen recording: create_ugc_video_overlay, a required corner position, overlay_size / overlay_scale for how large the inset sits. remove_black_background strips a black backdrop so you do not ship a black box.

That tool is an editor composite, not a generate. Iteration is the 480p preview pass; the charge is the export. The formula is editor previews and final export: free previews, one credit line per final render. Do not "save a shoot" by generating the inset.

A digital twin of the HRBP is a different act. HeyGen Avatar V5 exists as a catalog row for consented enrollment. An HRIS launch video is the wrong place to discover you did not have that grant. A likeness release that covers AI generation is the document if you will replicate someone. Most demo weeks should never open it. Overlay the founder you recorded, or stay faceless.

Captions beat a talking twin

Mute autoplay is how these files get watched. The screen already occupies the frame. A talking inset that restates the UI is optional. Captions that name the step are not.

Burn the line after picture lock. Do not prompt a video model to typeset "PTO request" on a fake laptop. Do not generate a second UI. The product in the recording is the identity hold. Motion that is not in the recording does not belong in the demo.

If you need a hook in front of the walkthrough, film three seconds of the founder pointing at the same screen. PiP that onto the recording. Do not invent an HRBP to "humanise payroll." Payroll is already human. The software is the proof.

FAQ

Can I generate a generic presenter if they are not a real employee?

A fully invented face is not a private-figure replica. On an HRIS demo it is still the wrong object: it reads as staff, it crowds the UI, and it spends a talking-head meter on a job captions already do. Record someone or stay on the screen.

Does picture-in-picture only work on slideshow exports?

No. Despite the slideshow_video_url parameter name, the overlay accepts any base video — including a product screen recording.

Should employees appear in the demo the way they appear in recruiting films?

Only if they volunteered, on camera, for this use. A recruiting consent is not a replica grant, and it is not a licence to generate them later. HR teams should treat a demo inset as a recorded overlay, not a twin.

Why not generate the whole UI so we do not show staging data?

Because a generated HRIS is a false product. Demo in a sandbox with fake names you control. Do not ask a model to invent the software. Identity for this category is the actual screen.