A sales asset library covers the combinations reps actually need: each key feature, framed for each buyer on the committee. With five features and six personas that's 30 clips. The part AI does well is the framing: a short persona-specific opener for every cell, generated on Kling v3 from a persona still made with Nano Banana Pro. The product itself should be a real screen recording. Thirty openers with persona stills, five feature shots and a few re-runs cost about 2,350 credits at 8frame's prices, roughly $23.50.
Correction, 2026-09-30: the first version of this page described a timed run with per-clip costs, 45 to 60-second Kling v3 clips and 30-second Seedance 2.0 clips (Kling v3 runs 3 to 15 seconds, Seedance 2.0 runs 4 to 15), 1080p output from both, a brand-pass step and a template on 8frame that don't exist, and generated "screen recording style" walkthroughs of a real product, which video models can't reproduce faithfully. This version keeps the matrix method and uses registry prices.
TL;DR
- Build the matrix first: features as rows, buying roles as columns, one framing line per cell.
- Real UI for the product: record your screen. A video model will invent UI and garble on-screen text; for sales assets, where prospects know or will see the real product, that costs credibility.
- AI for the framing: one persona still per role on Nano Banana Pro (21 credits), then a 5-second opener per cell on Kling v3 Standard (57 credits silent).
- 30 openers, 6 persona stills, 5 feature shots, 4 re-runs: 2,349 credits.
- Usage notes decide adoption. Write them the same day.
Why a matrix, not a pile of clips
Most SaaS asset libraries cover a fraction of the feature and persona combinations. A demo for a Head of Sales doesn't land with a DevOps lead; an enterprise walkthrough doesn't land with a founder. Reps either send the wrong clip or nothing. Defining the matrix before generating forces the framing decisions up front, so each cell gets its own angle instead of a swapped job title.
5-step build
Step 1: Feature inventory
List five to eight features, not twenty: the ones most often raised in discovery calls and demos. Typically the core workflow, a reporting surface, an integration, a collaboration feature and onboarding.
Write one sentence per feature: what the user does and what they get. "User connects the CRM from the integrations tab; contacts sync automatically." That sentence anchors every cell for that feature.
Step 2: Persona matrix
List the buying roles you actually meet: economic buyer (CFO or VP), champion, daily user, technical buyer, executive sponsor, and the skeptic who can kill the deal. Start from your last 20 won deals.
Features as rows, personas as columns. For each cell, one framing line: what this persona cares about and which objection the clip should answer. "CFO on the reporting dashboard: data freshness and audit trail, not visual design."
Step 3: Record the product, generate the framing
Record each feature once with a screen recorder, in your real UI and your real theme. Five features, five recordings, reused across personas.
Generate the persona layer on the canvas:
- One still per persona on Nano Banana Pro (21 credits each), used as the reference for every clip in that column so the same "CFO" appears throughout.
Chief financial officer in a glass-walled office, reviewing a report on a laptop, natural window light, neutral corporate interior, waist-up, 16:9, photorealistic, no text on screen
- One opener per cell on Kling v3 Standard from that still: 5 seconds, 57 credits silent.
The CFO leans toward the laptop, pauses, then nods slightly, camera slow push-in, calm office ambience, no text on screen
- Optional feature shots from real screenshots: a slow camera move over a screenshot on Kling v3 image-to-video. Check every frame for warped text; if it warps, do the move in your editor instead.
What to check: the persona looks like the same person across all five features, screens in the shot stay blank or blurred (no invented UI), and the framing line from the matrix is visible in the action (the CFO checking numbers, the DevOps lead at a terminal).
Step 4: Assemble and brand
Per cell: persona opener, the feature recording, an end card with the feature name. Do this in your editor with a template sequence so the 30 clips share type, colour and length. Add a 2-second title card at the start so reps can tell clips apart on a call. If you want voice, ElevenLabs text to speech on 8frame costs 13 credits per started 1,000 characters; a short persona-specific line per cell is 30 x 13 = 390.
Step 5: Sales handoff with usage notes
A library without usage notes gets ignored. Each clip should answer:
- Which deal stage? (Discovery, demo, follow-up, negotiation)
- Which objection does it answer?
- Where is it sent? (Inline in email, a Slack DM, embedded in a proposal)
Put it in one table that mirrors the matrix, one row per clip with a link, filterable by persona and stage.
What it costs
| Item | Model | Count | Credits |
|---|---|---|---|
| Persona stills | Nano Banana Pro | 6 | 126 |
| Persona openers, 5 s silent | Kling v3 Standard | 30 | 1,710 |
| Feature shots over real screenshots, 5 s silent | Kling v3 Standard | 5 | 285 |
| Re-runs (about 1 in 8 openers) | Kling v3 Standard | 4 | 228 |
| Total | 2,349, about $23.50 |
Add 390 for a voice line per cell. With Wan 3.0 at 720p (48 credits for 5 seconds) instead of Kling v3 for the openers, the 30 openers cost 1,440 instead of 1,710. If you use Seedance 2.0 with persona stills as references, frame them waist-up: Seedance rejects close-up face references with an E005 error.
Pitfalls
Generated UI. The fastest way to lose a technical buyer is a clip of your product that isn't your product. Keep generated footage away from the interface; record the real thing.
Over-templating. If every cell uses the same prompt with the job title swapped, the library reads as batch output. The matrix framing line is there to make each opener different.
Version drift. A library built in June shows a different product by September. Name every clip with the product release it shows (deal-room-cfo-v2.4.1) and review quarterly. Because the persona openers don't show UI, they survive a redesign; only the recordings need redoing.
FAQ
How do I decide which personas to include?
Start from your last 20 won deals. Who was on the buying committee, and who asked the most follow-up questions? Build for those people, not hypothetical buyers. Six roles covers most SaaS committees without making the matrix unmanageable.
Can AI generate the product walkthrough itself?
Not reliably for a live product: video models invent interface details and garble text. Use real screen recordings for the product and AI for the persona framing, openers and b-roll. A stylized AI stand-in only makes sense before launch, when there's nothing to record; see how to make a SaaS demo video with AI.
Can I update individual clips without rebuilding the library?
Yes. Each cell is independent. Keep the prompt and the persona still in the matrix doc; when a feature changes, re-record that feature and, if needed, re-run only the affected openers.
Sources
- 8frame Nano Banana, Kling v3, Wan 3.0, Seedance and ElevenLabs prices and lengths: the 8frame tool registry and content fact sheet, 2026-09-30
Matrix first, real product, generated framing. For more on SaaS video, see AI for SaaS companies. The 8frame canvas is free and unlimited; generation is paid from $19/month.