Aspect Ratios and Deliverable Formats for AI Video Explained

Which aspect ratios a project actually needs, why generating natively beats cropping, and how to plan a deliverable list before production.

Aspect ratio is treated as an export setting and is actually a compositional decision. A frame that is wide and one that is tall hold different amounts of information, place the subject differently and support different kinds of movement. Deciding the ratios at the end of a project means one of them was designed for and the others were derived, which is visible.

The four that cover almost all commercial work are the wide 16:9 for websites, presentations and events, the vertical 9:16 for full screen social placements, the square 1:1 for feeds, and increasingly the taller 4:5 which occupies more feed space than square without being full screen. A project intending to use more than one of these should compose for each rather than crop between them.

Cropping fails in a specific and recognisable way. A composition built with the subject on a third of a wide frame becomes a subject at the edge of a narrow one, and automated reframing chases whatever moves, producing the restless wandering that audiences identify immediately as repurposed content. The material is technically present and the composition is gone.

In generative production this problem is much cheaper to solve than in conventional shooting, which is a genuine advantage worth using. A shot can be generated at the target ratio rather than cropped from another, meaning each format can be a native composition at little additional cost. The requirement is that the storyboard specifies which formats are needed, because the framing decisions follow from that.

Where a single production must serve several ratios, the technique is to compose within a larger frame with the safe areas for each format marked. The subject and the essential action stay within the narrowest safe area, the wider formats gain context rather than losing the subject, and graphics sit on separate layers so they can be repositioned per format rather than rebuilt.

Text and graphics are what actually break during reformatting, more often than the picture. A lower third designed for a wide frame will overflow a vertical one, and text baked into a generated shot cannot be moved at all. Keeping every text element on its own layer, in a container designed with room to reflow, is what converts a format version into a re-render rather than a rebuild.

Event and large format work introduces ratios that no consumer platform uses. A wrapped LED stage may require an extremely wide canvas at a specific pixel count, and content must be built at that exact size because scaling costs sharpness. Wang et al. (2022), reviewing image quality measurement for large screen displays including LED, emphasise assessing quality from realistic audience viewing conditions, which for these formats means designing to the canvas rather than adapting a 16:9 master.

The safe area question differs per destination and should be checked against real interfaces rather than assumed. Platform elements occupy the bottom and top of vertical placements, captions occupy a band, and event screens may have physical obstructions. Content that is technically within frame and practically behind a play button is a common and avoidable delivery error.

The deliverable list should be written at quotation stage, not at delivery. A typical project needs a master, two or three social ratios, a silent version, subtitled versions per language, a thumbnail set and often an event specification version, which totals fifteen to twenty exports rather than one. Agreeing that list up front sets the price honestly and, more importantly, determines the framing decisions while they are still free.

The practical instruction for a client is to decide the destinations before the storyboard rather than after the edit. If the campaign will run vertically, the film should be composed vertically with a wide version derived, not the reverse. That single sequencing decision determines whether the social versions look designed or salvaged, and it costs nothing to make at the right moment.

References

Wang, J., Mou, X., & Wen, C. (2022). Measurement of image quality for large-screen displays based on audience viewing: Overview of standardization and preliminary study on laser projection displays and LED displays. Journal of the Society for Information Display, 30(6), 523–530. https://doi.org/10.1002/jsid.1095