LTXV Temporal Schedule Preview

LTXV Temporal Schedule Preview is the "look before you burn a render" node of the Temporal Director, and in LTX 2.3's world that's worth more than it sounds. One of the quiet annoyances of 2.3 is that the rebuilt VAE broke the old TinyVAE preview trick - you generate blind and find out at the end. A 10-second clip on a consumer card can be 5–15 minutes of sampling. This node renders a PNG of your entire schedule - segment durations, prompt text, LoRA envelopes, warnings - before the sampler ever runs. If your boundaries are wrong or a LoRA curve is in the wrong segment, you see it here in seconds instead of discovering it after a render.

It's the node that makes the Director's shared timeline feel real: it reads the same segment_data as everything else, so the picture is always of the exact schedule you're about to run.

How it works

It's a straight-up drawing node - it renders a PIL image on a dark background, seconds-first. The headline line is the total duration in seconds ("12.04 s"), with the technical details alongside: rendered frames, latent frames, FPS, segment count. Below that, each segment gets its own colored band on a timeline chart labeled by duration first ([1] 0.00–4.00 s · 96 frames), with the prompt excerpt and per-segment notes. Parser warnings and automatic normalizations show up inline, so if Schedule Sync had to recover missing pipe separators, you'll see exactly what it did.

The LoRA side is where the plot thickens: connect the data output of each scheduled Time-Gated LoRA to data_1 through data_4 and you get up to four colored envelopes overlaid on the same timeline, with incoming/outgoing transition markers. That's your q-curve check - you'll catch a mis-placed q_in_shift immediately.

Inputs and outputs that matter

• segment_data - from LTXV Schedule Sync. The only required wiring besides a preview output.
• data_1 … data_4 (optional) - envelope data from up to four LTXV Time-Gated LoRA (Scheduled) nodes.
• width / height - render size; defaults 1200×800, plenty.
• max_text_chars - how much prompt text per segment (default 54, up to 640). If your prompts are long, raise this and the height, since the text box height is the real limiter.

Outputs: image - feed it into a PreviewImage/SaveImage; report - the same information as a string, for text-display nodes or logging.

Installing and tips

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

Restart, hard-refresh. Also on ComfyUI Manager under "ComfyUI-LTXV-TimeGated-LoRA". No pip deps, and this node has no PromptRelay requirement - but it's only useful once you have a Director schedule wired up.

The habit to build: preview, check, then queue. Because LTX's prompt adherence is still its weakest axis, seed-hunting is the community's standard workflow - generate cheap candidates and pick - and this preview makes each candidate cheap by catching structural mistakes before the expensive part. One tip if the image looks cramped: bump max_text_chars to 120+ and raise the height; the chart area rescales to fit.