Conditioning (Set Area with Percentage for Video)

ConditioningSetAreaPercentageVideo is the video-aware version of the classic "Set Area with Percentage" node: it pins a piece of conditioning to a region of the latent - except here the region has a time axis too. Where the image version gives you x, y, width and height as percentages, this one adds temporal and z, so you can say "this prompt only applies to the left half and only for the first third of the clip." That's the whole point: video regional prompting isn't just spatial, it's temporal.

It's one of the quieter nodes in ComfyUI core, but if you've ever tried to keep two characters from bleeding attributes into each other across frames - or animate a background change that only happens halfway through - this is the built-in tool for it.

How it works

Under the hood it's a direct relative of ConditioningSetAreaPercentage: it writes an area tuple into the conditioning dict. The percentage flavor stores the box in 0–1 units (fractions of width/height/frames) rather than absolute pixels, and this video variant orders it as ("percentage", temporal, height, width, z, y, x). It also sets set_area_to_bounds to false, which matters: the area is respected as a region within the canvas, not expanded to the full latent. The strength value (default 1) scales how much this area conditioning influences the sample - the standard way to soften a region's grip when it's fighting the global prompt.

Inputs

• conditioning - the stream to confine.
• width / height - the region size as a fraction of the canvas (0–1, default 1 = full).
• x / y - the region's origin, also 0–1 fractions.
• temporal - the fraction of frames the condition applies to (0–1, default 1 = whole clip).
• z - the temporal start offset, also 0–1.
• strength - 0–10, default 1.

Output is the modified CONDITIONING. Chain a few of these with ConditioningCombine and each area keeps its own prompt.

Where it fits

Ships with ComfyUI core, sitting in the transform category of the video-model nodes. No model files involved. It's a niche-but-real tool: regional prompting in stills has been a steady community technique since the Latent Couple days, and video just adds the temporal axis nobody talks about. If you've already wrapped your head around image regional prompting, this is the same muscle with one more dimension.

Common issues

The classic error is thinking in pixels when everything here is a fraction - an x of 0.5 puts the region at the canvas's halfway point, not pixel 0.5. Second: the temporal axis is easy to ignore, but leaving temporal at 1 means the region holds for the whole clip, which may not be what you wanted for a mid-clip change. And with several stacked area conditionings, watch out for areas that overlap - the later-applied one wins in the blend, which is usually a surprise when it happens.