Choose explainer visuals by whether they document or illustrate a claim, compare stock and generated footage costs, and check rights before using either.
Direct answer: Use authentic recordings or verified screen demonstrations when the image is evidence of a real event, product or process. For clearly illustrative passages, choose licensed stock, a simple diagram or an explicitly identified generated clip according to accuracy, rights and review effort. My default is stock or diagrams for factual explainers; use generated realism only when it adds a clear explanatory benefit without implying that an invented scene actually happened.
The important distinction is not simply whether an image is real. Genuine footage of the wrong factory, device or location can mislead just as readily as a synthetic scene. Viewers infer meaning from where you place the image and what the narration says over it.
Start by deciding what each visual is allowed to claim. That prevents an attractive clip from becoming accidental evidence for something it does not document.
Establish the visual evidence boundary
The visual evidence boundary is an editorial method for separating documentation from illustration before choosing assets. Mark each planned segment as evidence of a specific thing or as an explanation of a general concept.
This is a decision within a practical AI video workflow. It does not assume that adding AI improves the result. An accurate diagram or a recording made with equipment you already own may be the better option.
Before downloading or generating assets, identify the intended publication, audience, territories and reuse. Check the relevant licence, agreements and any restrictions involving people, brands or locations. Do not upload confidential reference material or another person's likeness to a generation service without appropriate authorisation and verified handling terms.
For UK use, government guidance identifies permission, ownership or an applicable copyright exception as the basis for using protected material. A file being available online does not settle the question. Obtain qualified local advice for a consequential uncertainty; jurisdictions and contracts differ. UK guidance on using copyright material.
Ask what the viewer will infer
For each visual, finish the sentence: “A viewer could reasonably think this shows...” If the answer names a real incident, product behaviour or person, you need evidence that the footage actually represents it.
A stock shot of a server room can illustrate computing infrastructure. Placing it under a claim about a named provider's security incident may imply that it shows the affected facility. The clip's licence does not make that implication accurate.
A generated view of a mechanism may help explain a concept if its relevant relationships are correct and its illustrative status is clear. It should not stand in for a real product demonstration, a recovered historical scene or proof that a particular event occurred.
The Society of Professional Journalists' ethics code warns against distorted visual context and calls for clear identification of illustrations and re-enactments. This is an editorial principle, not a universal legal disclosure rule. SPJ Code of Ethics.
Compare the options by the job they must do
| Visual option | Evidential role | Review requirement | Cost or dependency to inspect |
|---|---|---|---|
| Authentic recording | Can document the specific event or process shown | Verify identity, context and edits | Access, filming, permissions and storage |
| Licensed stock | Usually general illustration unless provenance matches the claim | Check scene accuracy and implied context | Licence scope and asset fees |
| Generated clip | Illustration of an invented scene or concept | Inspect factual details and disclosure needs | Generation, rejected outputs and service terms |
| Diagram or simple animation | Explicit explanation of relationships | Verify labels, sequence and scale assumptions | Design and revision effort |
My recommendation favours diagrams when the information is conceptual and authentic recording when the information is observational. Stock can supply useful atmosphere or general context. Generated clips should earn their place by explaining something more clearly, not merely by looking expensive.
The strongest case for generated illustration is a visual that would otherwise be impractical to create and whose purpose is clearly conceptual. The condition is that you can check the relevant details and keep the audience from mistaking it for documentation.
Inspect the asset at the point of use
Review the clip with the actual narration, title and surrounding shots. A generic image can acquire a specific implication through an adjacent sentence. Label illustration where needed at the point the inference arises, not only in a distant production note.
For stock, verify what you can about the scene and its provenance. Do not assume that a search keyword establishes the device model, location or activity. If the detail matters and cannot be confirmed, use another asset or change the claim.
For generated material, inspect the mechanism, interfaces, signs, labels, proportions and sequence of action relevant to the explanation. Do not accept a technically impossible demonstration because the lighting looks convincing. If the concept can be explained with fewer, checkable elements, simplify it.
Keep a record of the selected asset, source, licence or service terms checked, permitted use and required credit. A later editor needs that information before adapting the video for another platform or campaign.
Calculate the cost of ten visual segments
Imagine an explainer needing ten segments: three authentic demonstrations, four illustrative passages and three diagrams. All prices and timings below are hypothetical planning assumptions, not vendor quotes or measured production results.
Assume the three authentic segments take 15 minutes each to prepare and check, and the three diagrams take 20 minutes each. Shared effort is 3 × 15 + 3 × 20 = 45 + 60 = 105 minutes, whichever route supplies the four illustrative passages.
For stock, assume four licences at £18 each and 12 minutes per asset for selection, licence inspection and contextual review. Cash cost is 4 × £18 = £72. Additional effort is 4 × 12 = 48 minutes. Total production effort under these assumptions is 105 + 48 = 153 minutes.
For generated clips, assume three candidates per usable passage, giving 4 × 3 = 12 candidates. At an illustrative £1 charge per candidate, generation costs £12. Assume eight minutes of preparation and review per candidate, plus ten minutes of final context and permission checking for each selected clip.
Generated-route effort is 12 × 8 + 4 × 10 = 96 + 40 = 136 minutes, or 105 + 136 = 241 minutes including the shared work. Compared with stock, the fictional cash difference is £72 − £12 = £60, but active effort increases by 241 − 153 = 88 minutes.
That does not prove stock is cheaper or AI is slower in general. It shows why fees alone do not settle the decision. Substitute actual billing, rejected-output counts, review effort and licence conditions for your project. Do not convert those extra minutes into a wage cost unless that reflects a real expense or opportunity you can justify.
Choose a stopping condition for generated attempts
Before generating, define the factual details the clip must preserve and an effort budget. If repeated outputs fail the same critical requirement, switch to a diagram, stock asset or authentic recording rather than lowering the acceptance standard.
For stock searches, use a similar stopping condition. If no asset matches the actual process, do not select a vaguely related clip and let the narration conceal the mismatch. A simple labelled visual can be more accurate and less distracting.
Keep rejected candidates separate from approved assets so a later editor does not accidentally use them. Preserve originals and the licence record for accepted material according to the project's retention requirements.
Decide the visual treatment before producing the full video
- Spend ten minutes marking every planned segment as documentation or illustration.
- Use 15 minutes to identify the factual details and rights questions that determine each choice.
- Test one difficult illustrative segment with two credible approaches, including a simple non-AI option.
- Compare the result in the actual edit, then choose the route with acceptable meaning, permissions and total effort.
Stop when a visual implies evidence you do not have or its rights cannot be established. Replace the asset or change the presentation before publication. An attractive image is optional; an accurate explanation is not.
Related guides
Frequently asked questions
Is stock footage automatically more trustworthy than AI-generated footage?
No. Stock may depict a real scene, but it can still be the wrong scene for your claim. Check whether the device, location, people and activity match what the edit implies. A genuine laboratory shot does not prove that a named product was tested there. Generated material has a different status because the scene is invented, but both require contextual review. For factual evidence, use material whose provenance and relevance you can establish. For illustration, make that role clear and choose the asset that explains the idea accurately rather than ranking trustworthiness solely by how the image was produced.
Can I use generated footage to demonstrate a real software feature?
Not as proof of the feature's actual behaviour. Use a verified recording of the relevant product and account conditions for a factual demonstration. A generated interface can invent buttons, responses or steps that viewers cannot reproduce. If you use a conceptual mock-up, identify it clearly and avoid presenting its details as current instructions. The exception is a deliberately illustrative explanation where the exact interface is irrelevant and the relationship shown is accurate. Even then, review the final edit for misleading implications. A realistic-looking screen is not evidence that a product works as depicted.
Does labelling a clip as AI-generated solve every problem?
No. A label can clarify origin, but it does not correct a false mechanism, misleading context, missing permission or inappropriate portrayal of a person. The visual still needs to support the explanation accurately. Place essential identification where viewers will encounter it with the clip, rather than relying only on a distant note. Follow the relevant platform, agreement and jurisdictional requirements, seeking qualified advice where needed. This article's recommendation is editorial, not a universal legal disclosure standard. If the asset remains misleading after identification, change or remove it instead of treating disclosure as a complete remedy.
Should I always pay for stock rather than use free assets?
No. Price alone does not establish licence suitability, provenance or visual quality. A free asset with clear terms may fit your use, while a paid asset may carry restrictions that do not. Check the actual licence, required attribution, permitted distribution and any relevant releases or limitations. Keep a record of the terms you relied on. Do not assume that “free download” means unrestricted commercial reuse or that payment settles every rights question. The right choice is the asset that accurately serves the explanation under a permission arrangement you understand, at a total cost justified by the project.
When is a diagram better than either type of footage?
Use a diagram when the important information is a relationship, sequence or distinction that realistic footage would obscure. A clear labelled flow can explain where information goes more precisely than a generic office scene. Verify the labels and assumptions, and avoid implying exact scale or timing when the diagram is schematic. Footage is preferable when viewers need to observe an actual action or recognise a real object. You can combine the approaches, using a demonstration for evidence and a diagram for explanation. Do not choose moving imagery merely because a static, simpler visual seems less fashionable.
How should I prepare assets for future reuse?
Keep the source identity, approved version, licence or relevant service terms, required credits and intended use alongside each accepted asset. Record whether it is documentation or illustration and any contextual restrictions the next editor needs to preserve. Do not assume permission for one publication automatically covers another purpose, territory or distribution arrangement. Before reuse, recheck the applicable agreement and the new edit's implications. Keep sensitive source material in its approved location. A well-labelled asset library should help a future editor understand what the visual can legitimately and accurately do, not merely where the file is stored.
Sources and verification
- SPJ Code of Ethics, consulted 11 September 2026 for visual context and identification of illustrative material.
- UK government: using copyright material, checked for the general permission framework. No specific asset licence or legal outcome is assumed.
- The parent was read locally because its public route was unavailable. The visual evidence method, fees and production timings are illustrative editorial assumptions.
This article is practical guidance. Apply it in proportion to your tools, evidence, risks, and responsibilities.
