SaaS demo from a 48-second recording: trim, crop and cues for $0.24

A 48-second screen recording becomes a tighter SaaS demo clip on Sume for $0.24: video trim to cut the lead-in ($0.02), video filter to crop to the part of the screen that matters ($0.02), and the standalone caption job to burn your own cue text ($0.20). No model generation is involved, so the UI in the clip is your real product.

Why use the recording and not a generated demo

A generated video of software invents buttons. A recording shows the real interface, and the three steps here only cut, crop and label it. That keeps the demo accurate, which matters when a prospect compares the clip with the product.

All three jobs read a clip that already lives on media.sume.com in your workspace, so import the recording first with the media import endpoint.

The three jobs

Trim takes start and duration. If the first 4 seconds are a cursor moving to a window, send start 4 and duration 44. Exact precision re-encodes frame-accurately. Crop is an op in video filter, given as fractions of the frame: x and y in 0 to 1, width and height in 0.05 to 1. For example x 0.08, y 0.12, width 0.84, height 0.78 keeps the app window and drops the dock and the menu bar. The compiler rounds to even values for yuv420p. The filter accepts sources up to 300 seconds, and POST check is free before you encode.

The caption job needs a public HTTPS video URL, and with cues it does not run speech-to-text. Each cue has text, start and end in seconds, so a silent recording gets labels such as step names at the right moments.

Caveats

Crop changes the aspect ratio. A 16:10 recording cropped to 0.84 by 0.78 of the frame is no longer 16:10, so check the result with video inspect before you pick a platform. Caption styles for Latin text default to slam, which is loud for a demo; choose black-outline or a calmer style and use the design overrides for color.

Sume bills each job separately and the filter and trim jobs have no provider inference, only worker ffmpeg. If a cue timing is off, restyle or correct it by sending source_caption_id with new words rather than redoing the whole chain.

A practical note

Keep each cue short, at most about six words, and time it to the action: the label should appear as the cursor reaches the control and leave when the screen changes. If a clip has narration you want to keep, do not use cues; send the clip without them and let speech-to-text produce the captions, or provide script_text to match your product terms exactly. Brand names and feature names are the words speech-to-text most often misspells, so for a demo the script route is usually the safer one.