Reframe API use cases

The Reframe API modifies a video's aspect ratio by intelligently tracking the most important part of each shot and recomposing the frame around it. All content in the reframed output is derived solely from the source video, and the API never generates new pixels to fill space.

Best use cases

The Reframe API performs reliably on:

• Videos requiring standard social media aspect ratios: Conversion of videos with a 16:9 aspect ratio to 9:16, 1:1, 4:3, and 4:5 outputs for different channels is the main use case. You can also specify an exact output resolution (for example, 1920×1080) for pixel-perfect delivery. 
• Live-action and real-world footage: Interviews, product shots, lifestyle and event footage, and other camera-captured videos reframe cleanly. Natural camera motion is automatically smoothed out, without any visible artifacts. 
• Shots featuring one dominant subject: The dominant subject, such as a person, product, or focal point, stays in frame throughout the recomposition. With Semantic Subject Lock, you can name the target subject using a keyword or prompt (for example, man in a yellow jacket or frisbee), and the Reframe API will track it across every shot. 
• Multi-shot videos with clean cuts: Scene-edit detection, when activated, allows the API to analyze each transition and hold a stable composition within every shot across all output ratios. 
• Videos that need consistent overlays: Pre-generated overlays (PNG or GIF) can be applied with control over timing, position, scale, and looping, and stay aligned across all aspect ratios you generate.

Poor-fit use cases

Because the Reframe API only repositions and crops within existing pixels, it never generates new content to fill space. The following content types will degrade or fail when reframed:

• Full-frame on-screen text and lower-thirds: Titles, captions, credits, and graphics burned into the source video occupy the full width of the original aspect ratio. When the frame is cropped to a narrower ratio, such as from 16:9 to 9:16, the Reframe API truncates the text at the edges because it cannot reflow burned-in content—it can only move the crop window. For videos with text and graphics, reframe clean videos first, and then use DGR (with different-sized MOGRTs) to overlay text on them.
• Multi-subject scenes with simultaneous action: Semantic Subject Lock can track only one subject throughout the recomposition. When two or more subjects are equally important and spatially separated, such as a two-person interview in a wide shot, a panel, or a sports play with a ball and a distant player, the crop can only hold one subject. The others fall outside the frame.
• Rapid or ambiguous scene cuts: Scene-edit detection maintains compositional integrity across transitions, but very fast cuts, hard flash transitions, or match cuts can cause the subject-tracker to lock on to the wrong element, producing jumps in the crop path. 
• Fixed-composition graphics content: Screen recordings, slide decks, dashboards, and footage with extensive motion graphics have no trackable subject and no safe crop. The Reframe API treats every pixel in such content as important, and so such content should not be sent to the API.
• Animation and motion graphics: Animated content, such as logo builds, kinetic typography, title sequences, and brand animations, doesn't reframe reliably. Such content often features crisp graphic edges and fast-moving elements that require precise, frame-by-frame framing, and it frequently combines multiple elements moving in different directions at once. The Reframe API is fine-tuned for real-world footage and cannot track animated elements precisely enough, so the output may look slightly off or lag the motion. For animation, render each aspect ratio from the original animation source, or its MOGRT through DGR, rather than reframing a finished export. 
• Unsupported outputs: The Reframe API always produces opaque H.264 or HEVC output. Any transparent background (an alpha channel) is flattened, as alpha isn't preserved. ProRes is accepted as input for decoding, but not produced as output.