video-constraint-map — Traditional AI Model
Video Constraint Map measures a long source once and reports what a planner cannot infer from a single frame: how deep a bottom crop may go before it eats the subject, read off the source's own burned-in captions.
Details
- CategoryTraditional
- Year-1
- License
Inputs & Outputs
- VideoInputvideo
The source video to measure (before trimming)
- SegmentsInputsegmentsoptional
Optional candidate segments to restrict measurement to. Leave unconnected to measure the whole source, which costs proportionally more.
- ConstraintsOutputstring
JSON: derived crop depth, caption spans, cut-boundary spans, and the per-window subject grid, each with its evidence
Parameters
- sample_fpsfloatdefault 1range 0.5–4
How many frames per second to measure. Higher resolves caption and subject timing more finely and costs proportionally more.
Measurement sample rate. Every sampled frame costs one detector pass over the whole frame plus one per candidate window, and the detector is the entire runtime — so doubling this doubles the run. 1/s (default) resolves caption boundaries to within a second; raise it to 2/s when cut timing matters more than run time.
- pan_positionsintegerdefault 5range 3–9
How many candidate horizontal window positions to measure per frame. More positions give the planner finer framing choices and cost proportionally more.
Number of candidate 9:16 window positions measured per sampled frame, spread evenly across the width that the window can reach without being clamped. Reported as window CENTRES in fractions of source width, which is exactly what the Reframe node's pan_from / pan_to take. 5 (default) steps the window about a third of its own width at a time on a 16:9 source; 9 is the finest and costs nearly twice as much.
- subjectstringdefault Autooptions: Auto, Person, Pet (cat/dog), Animal
What counts as the subject when measuring. Auto considers people and animals together; naming one stops the others competing.
Which subject the measurement looks for, using the same classes as the Reframe node so the two agree. 'Auto' (default) considers people and common animals together, which is wrong when a bigger, nearer animal shares the shot. Name the subject to remove that competition: 'Person', 'Pet (cat/dog)', or 'Animal' (animals only, no people).
- bottom_safe_fracfloatdefault 0.1667range 0.05–0.33advanced
A subject reaching into this bottom fraction of the frame turns the caption crop off for that moment, so the crop cannot cut it away.
How much of the frame's bottom counts as at risk from the caption crop. When a subject's box reaches into it, the map reports the crop as off for that moment and the planner keeps full height there. The default sixth comes from a dog filmed from behind, whose tail sits in the bottom sixth and was repeatedly cropped away. Raise it to protect more, at the cost of keeping the source's captions on screen more often.
- max_samplesintegerdefault 400range 1–2000advanced
Upper bound on measured frames. Over the limit the sample grid is thinned evenly rather than cut short, so the end of the video is still measured.
Ceiling on how many frames are measured, so a long source cannot turn into an unbounded run. When the requested sample rate would exceed it the grid is thinned uniformly across the whole timeline — never truncated, which would silently stop measuring the end — and the output says so in its warnings.
Tags
Alternatives in Video Editing
- Audio Edit
Trim a sound, fit it to a video or a length, dip or mute parts, and fade it.
- Audio Finish
Normalize a video's loudness to a broadcast target (2-pass loudnorm + peak limiter).
- Audio Mix
Mix narration, music and sound effects into one track, with ducking and loudness.
- Audio Replace / Mix
Replace or mix a clip's audio with a new sound, keeping the video as is.
- Chroma Compose
Cut a green-screen person out of one video and place them over another.
- Hook Teaser (Cold Open)
Prepend a cold-open teaser of each clip's hook moment to the front.
- Ken Burns (Still to Video)
Animate a still image into a video with a moving crop window.
- Narration Mux (Cold Open)
Prepend a voiceover cold-open (frozen first frame + narration) to a clip.