Higgsfield Genjutsu example pair, measured: a shorter output, and a silent source with nothing to keep
Measurements of the example source and output videos shipped with Comfy-Org's pinned Genjutsu motion-transfer template. The output has 89 frames to the source's 96 (both 24 fps). The source audio decodes to digital silence while the output's does not. The output's embedded request names the template's model, resolution, seed and exact reference images. Nothing was played or listened to.
What this is: measurements from a private, archived analysis of the example source and output videos and reference images shipped with Comfy-Org workflow_templates for the Higgsfield Genjutsu motion-transfer template. The inputs were ffprobe frame tables, a loudness-envelope decode of the first audio stream, and container tags. The example media are not reproduced here. No video was played and no audio was listened to.
Length: the source clip has 96 decoded frames (4.000 s container duration) and the output has 89 (3.709 s), both reported at 24 fps. The template's note says output length follows the source clip, but in this example pair the output is 7 frames (about 0.29 s) shorter.
Audio: at the analysis decode (8 kHz mono, 16-bit samples), every sample of the source clip's audio track is exactly zero in all 40 analysis windows: digital silence. The output's track is not silent (about -35 to -14 dBFS per window). So the output's audio is not the source audio, although the template note says audio is kept. What the output audio contains was not established. An earlier automatic similarity label for this pair was invalid, because silence gives no correlation; it is corrected to not computable.
Embedded request: the output container carries an editable ComfyUI prompt tag. It names model genjutsu-v1.0, 720p and seed 42, as the template does, and its reference-image change hashes equal the SHA-256 of the two reference PNGs shipped with the template. Its prompt wording differs from the prompt stored in the template. Container tags can be edited, so this is a consistency check, not provenance proof.
For users: check length and audio on your own outputs rather than relying on the template note. If the source has no audible audio, there is nothing to keep.
Not covered: visual likeness, costume or setting comparison, motion or camera fidelity, and quality. Example-media rights are unresolved and the media stays private.
Structured fields
{
"citations": [
{
"url": "https://github.com/Comfy-Org/workflow_templates/tree/0cc0fba562efe9f68e48050c2b82ed6495050bc2",
"note": "Pinned snapshot holding the measured example pair; exact derived hashes and byte ranges retained in source pins."
}
]
}Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| documentation | unknown | permitted | permitted | permitted | unknown | Example media remain private: MIT scope for media unresolved, likeness and input-origin terms unreviewed. The embedded prompt is not quoted. |