Combining Live Footage With AI Generated Shots

How to cut filmed and generated material together so the joins disappear, what has to match, and where to place the transitions.

Most good commercial films now contain both filmed and generated material, and the quality of the result depends less on either source than on whether they belong to one another. A film that cuts visibly between real and constructed material undermines both: the filmed shots look flat by comparison and the generated shots look artificial by contrast. Making the joins disappear is a set of specific technical decisions, not a matter of taste.

The first decision is made at storyboard stage, and it is where the transitions sit. The safest places to move between sources are on a change of scene, behind a wipe or an object crossing frame, on a hard cut with a change of angle and subject, or across a moment of motion blur. The riskiest is a straight cut between two similar framings of the same subject, which invites direct comparison.

Lighting direction is the property that must match and the one most often overlooked. If filmed material has a key light from the left and the generated shot places it on the right, the sequence reads as wrong even when the viewer cannot name why. Deciding the lighting direction at the board, and specifying it in the generation prompt as well as on the shoot, is what keeps a hybrid sequence coherent.

Lens character is the second. Filmed material carries a specific focal length, depth of field, distortion and bokeh, and generated material has its own default look that is frequently wider and deeper than a cinema lens. Specifying the lens character in generation, and matching the depth of field in post where necessary, closes most of the gap. Adding subtle optical imperfection to generated shots, a little vignetting and chromatic aberration, brings them toward the filmed material rather than the reverse.

Colour and grain are the third, and they are handled in finishing rather than in production. The whole timeline should be corrected to a common base before any look is applied, and grain should be added uniformly at the end, after grading, so that filmed and generated shots share the same texture. A sequence where some shots are clean and others carry sensor noise is immediately readable as mixed sourcing.

Camera movement is the property that most reliably gives generated shots away when it is wrong. Filmed movement has weight: a handheld shot has micro corrections, a dolly has inertia, a crane has a slight settle at the end. Generated movement is often uniformly smooth, which reads as artificial next to filmed material. Adding subtle instability to generated moves, or matching the generated shot to a locked off filmed shot rather than a moving one, both work.

Compositing is what allows the best of both to appear in one frame, and it is where brand critical elements belong. A generated environment with a filmed product, or a filmed presenter against a constructed background, or an accurate 3D product inside a generated world, are all standard constructions. The rule that governs them is the same one that governs the whole category: anything requiring exactness is placed, not generated.

There is a disclosure and credibility dimension that should be settled rather than left implicit. Kirk and Givi (2025) found that perceptions of AI authorship shape consumer responses to marketing communications and can produce negative reactions in some conditions, and Farooq and de Vreese (2026) documented how awareness of AI generation affects authenticity judgements. In practice the safe division is that anything functioning as evidence, real people, real premises, real results, is filmed, while environments, atmosphere and illustration may be constructed. A hybrid film built on that division does not have a disclosure problem.

Nguyen et al. (2026), in a cross national study of responses to AI generated advertising, found that viewer responses depend heavily on execution and perceived authenticity rather than on the technology itself. That is the empirical basis for the effort described here: audiences are not reacting to whether generation was used, they are reacting to whether the result holds together.

The practical test before delivery is to watch the film with someone who does not know how it was made and ask them to identify which shots were filmed. If they cannot, the integration worked. If they can, the difference is almost always one of the four properties above, lighting direction, lens character, colour and grain, or movement quality, and it is fixable in finishing rather than requiring anything to be remade.

References

Kirk, C. P., & Givi, J. (2025). The AI-authorship effect: Understanding authenticity, moral disgust, and consumer responses to AI-generated marketing communications. Journal of Business Research, 186, Article 114984. https://doi.org/10.1016/j.jbusres.2024.114984

Farooq, A., & de Vreese, C. (2026). Deciphering authenticity in the age of AI: How AI-generated disinformation images and AI detection tools influence judgements of authenticity. AI & Society, 41(1), 493–504. https://doi.org/10.1007/s00146-025-02416-5

Nguyen, K. M., Phan, T. M., Tran, Y. N. N., Nguyen, A. T., Nguyen, T. L. N., Hoang, G. H., Tran, T. T., & Nguyen, N. T. (2026). Evaluating the efficacy of AI-generated advertising: A cross-national analysis of customer responses on brand perceptions and customer engagement with evidence from Vietnam and Australia. Journal of Global Scholars of Marketing Science, 36(2), 293–341. https://doi.org/10.1080/21639159.2026.2617659