Nodes/ComfyUI-LTXV-TimeGated-LoRA/LTXV Temporal Envelope Inspector (v1.3.0)
ComfyUI Node

LTXV Temporal Envelope Inspector (v1.3.0)

Check LTX 2.3 timing and envelopes without touching the model

By Jinx138·Created 3 months ago·Updated 27 days ago· 21
LTXV Temporal Envelope Inspector (v1.3.0)
  • video_latent
  • video_latent
  • report_md
  • report_json
  • profile_plot
  • data
envelope_modelocal
effect_regionmiddle third
strength_before0.00
strength_during1.00
strength_after0.00
outside_strength0.00
ramp_modeq_curve
q_in1.00
q_out1.00
transition_frames0

The expensive thing in LTX 2.3 work isn't the node - it's finding out your schedule was wrong after a ten-minute render. The Temporal Envelope Inspector is this pack's answer to that: a non-destructive audit node that resolves your video timeline and computes the exact strength envelope a Time-Gated LoRA would apply, without loading a LoRA or patching the model.

It ships in the same Jinx138 pack as the LTXV Time-Gated LoRA (LTX 2.3) and the CPU curve preview, and it's the middle of a three-node diagnostic chain: plan the envelope, inspect it, preview the plot, then commit to the real render. The core idea is that a timeline costs almost nothing to compute, so you verify before you spend.

How it works

Feed it the final video-only video_latent and it resolves the real rendered frame count - LTX 2.3's temporal compression is fixed at 8, so the node derives the timeline from the latent's temporal dimension - then builds the envelope your settings describe. Crucially, it uses the same timing engine as the productive node, so what it reports is what the Time-Gated LoRA will actually do. That engine match is the whole reason it exists.

Inputs and outputs

Inputs mirror the main node: envelope_mode (transition_frames, v1.1_local, or v1.1_hold_strength), effect_region for the active interval, strength_before / strength_during / strength_after, ramp_mode and ramp_q for the transition shape, and transition_frames for the legacy crossfade mode. outside_strength is the one extra - only used in v1.1_local mode, it sets the level in the timeline outside the immediate before/active/after envelope. None of it touches your model.

Outputs: video_latent (passed through unchanged, so you can drop the Inspector inline without rerouting anything), report_md (human-readable), report_json (for automation), profile_plot (an IMAGE of the frame vs latent profile), and data - the same LTXV_ENVELOPE_DATA payload the CPU curve preview consumes. The README's recommended wiring:

video_latent
  -> LTXV Temporal Envelope Inspector
       data      -> LTXV Envelope Curve Preview (CPU) -> Preview Image
       report_md -> Show Text / PreviewAny

You get both the numbers and the picture, then you either remove the Inspector before the real run or leave it inline - it's harmless either way.

When it's actually worth it

Three cases. One: you're hand-planning segments (the manual_frames style of scheduling) and want exact region boundaries in rendered frames rather than percentages. Two: you've stacked multiple gated LoRAs and want to confirm their regions don't accidentally overlap or leave a gap. Three: you suspect the latent feeding the main node is wrong - the Inspector makes the mismatch visible immediately, which beats debugging it after a failed sample.

Install and caveats

Install is shared with the pack, and it's the easy kind: no dependencies, no model files, just the clone.

cd ComfyUI/custom_nodes
git clone https://github.com/Jinx138/ComfyUI-LTXV-TimeGated-LoRA

Restart and look under LTXV/Diagnostics (or ComfyUI Manager → "LTXV Time-Gated LoRA").

Caveats, because this is a young pack - first public node release, no community threads about it. The class name says v1.1, and the repo has moved on since: newer builds split the q-curve into separate in/out controls and add a much bigger Temporal Director workflow. The comfy.icu page reflects the v1.1 schema, so if your widgets look slightly different after an update, that's version drift, not you. And remember the Inspector reports the envelope you configured, not whether that envelope is a good idea - a perfectly resolved schedule can still flicker if the LoRA is a stochastic detail type like rain or grain. Use it to catch timing mistakes, not to predict quality.

CategoryLTXV/Diagnostics

Inputs (11)

NameTypeDefaultDescription
video_latentLATENTFinal video-only LTX LATENT used only for timing inspection. Connect before audio/video concat. This node passes it through unchanged.
envelope_modeCOMBOlocallocal and hold_strength use segment-length q-curve envelopes. transition_frames reproduces the legacy boundary-crossfade model.
effect_regionCOMBOmiddle thirdActive region controlled by strength_during. The inspector resolves exact frame boundaries from the video_latent timeline.
strength_beforeFLOAT0.00-8–8Target LoRA strength before the active region. In v1.1_local this means the immediate preceding neighbor segment; in hold_strength it is also held earlier.
strength_duringFLOAT1.00-8–8Target LoRA strength inside the selected active region.
strength_afterFLOAT0.00-8–8Target LoRA strength after the active region. In v1.1_local this means the immediate following neighbor segment; in hold_strength it is also held later.
outside_strengthFLOAT0.00-8–8Used only in v1.1_local. Strength outside the immediate before/active/after envelope area. Ignored by v1.0rc1 and hold_strength modes.
ramp_modeCOMBOq_curveflat uses hard segment strengths. q_curve uses q_in/q_out power curves over the neighboring segment length.
q_inFLOAT1.000.01–16Incoming q_curve: strength_before -> strength_during.
q_outFLOAT1.000.01–16Outgoing q_curve: strength_during -> strength_after.
transition_framesINT00–4096Used only by transition_frames mode. Ignored in v1.1 envelope modes, where the segment itself is the ramp duration.

Outputs (5)

NameTypeDescription
video_latentLATENT
report_mdSTRING
report_jsonSTRING
profile_plotIMAGE
dataLTXV_ENVELOPE_DATA