LTX US Config (CRT)

The pack's LTX 2.3 Unified Sampler is a monster node that runs the whole generation pipeline, but it deliberately doesn't hold the creative inputs. That's what this node is for. LTX 2.3 US Config is the "what are we making" half of the pair: your prompt, your seed, your first frame, your reference video, your source audio. The sampler is the "how" - this node is the "what," and it hands the whole bundle downstream as a single config_pipe.

Separating them is a genuinely smart design for a workflow this big. You can keep the config node stable while you swap sampler settings, or keep several config nodes (different prompts, different first frames) feeding one sampler to A/B ideas.

How it works

Two required inputs, and a pile of optional ones that only matter in specific modes:

• prompt - the text-to-video / description of what you want. For I2V and V2V, prompts still steer motion and style; for T2V it's the whole ballgame.
• seed - the reproducibility knob.

The optional inputs are the interesting part:

• Image I2V / V2V FirstFrame - the starting image for image-to-video, or the anchor frame for video-to-video.
• Video (V2V image batch) - the source footage for video-to-video mode, fed as an image batch.
• V2V Depth (override) - plug in a precomputed depth map and it skips the DepthAnything inference entirely. That's a real time-saver if you've already got depth from another pass; leave it empty to let the pipeline compute it.
• source_audio - the audio track, which enables synchronized-audio generation and drives frame count if the sampler's frame_count_from_audio is on.
• Then a set of overrides that all default to 0/off, meaning "use the sampler's value": MegaPixels, Frames, Width T2V, Height T2V, HQ. Want this run at 2 MP while the sampler defaults to 1? Set the override here instead of touching the sampler.

Output is a single config_pipe (LTX23_US_CONFIG_PIPE) - that's what wires into the Unified Sampler's config_pipe input.

Installing it

Same pack: ComfyUI Manager → CRT-Nodes, or clone + pip install -r requirements.txt, restart. It lives under CRT/LTX2.3, right next to the sampler.

Where people get burned

The overrides default to 0, which reads as "zero megapixels" - it isn't. Zero means "unset, use the sampler's value," and treating it as a real zero is the classic confusion. Also, mode mismatches bite: a T2V run ignores your first frame, a V2V run needs the video batch, and if you connect a depth override that doesn't match the video dimensions you'll get conditioning chaos. Keep the pieces of the graph consistent with the sampler's workflow_mode dropdown and this node stays boring - which is exactly what you want from a config node.