🅛🅣🅧 Guider Parameters

Guider Parameters is a small config node that packages up one set of guidance settings - CFG, STG, and a few LTX-2-specific knobs - into a bundle you hand to the Multimodal Guider. On its own it does nothing visible; its whole job is to be the tunable "settings card" that the guider reads. The clever part is that it's chainable: you can stack one Guider Parameters for the video modality and another for audio, and the Multimodal Guider applies each to the right stream.

It's part of ComfyUI-LTXVideo, Lightricks' official pack, and it's part of the January 2026 "Multimodal Guider" drop - the release that split LTX-2's prompt adherence, temporal stability, and audio-video sync into independently tunable controls. Lightricks' own framing was "no more choosing between 'follows the prompt' and 'doesn't fall apart.'" This node is where you set those trade-offs.

How it works

You configure one modality's guidance here and output a GUIDER_PARAMETERS object. Because LTX-2 generates video and audio in the same model, the two streams often want different guidance - so you make a video parameter set and an audio parameter set, and chain them together through the optional parameters input. The Multimodal Guider then reads the combined bundle and steers each modality accordingly.

The inputs that matter

• modality - VIDEO or AUDIO. This is which stream the settings apply to. Make one node per modality.
• cfg - classifier-free guidance strength (default 1, correct for distilled LTX).
• stg - Spatiotemporal Skip Guidance strength (default 1), the dial that sharpens motion/structure.
• modality_scale - the cross-modality control (default 0), governing how strongly audio and video pull on each other - the audio-video sync knob that made this drop notable.
• perturb_attn / cross_attn - toggles (both on by default) for the attention-perturbation and cross-attention paths STG uses.
• rescale and skip_step round out the set (guidance normalization and a step to skip).

The optional parameters input takes another Guider Parameters output, so you can chain video + audio into one bundle. The single output is GUIDER_PARAMETERS → feed it into Multimodal Guider's parameters socket.

Installing it

ComfyUI Manager → Install Custom Nodes → search LTXVideo → install → restart, or cd ComfyUI/custom_nodes && git clone https://github.com/Lightricks/ComfyUI-LTXVideo and restart. It appears under lightricks/LTXV. It's pure config - no weights of its own - but it only makes sense inside an LTX-2/2.3 audio-video graph, which needs the LTX checkpoint and Gemma text encoder downloaded.

Where people get burned

• Expecting output from this node alone. It produces a settings object, not a guider. Nothing happens until it's plugged into a Multimodal Guider. If your graph "does nothing," check that the chain actually terminates at the guider and then a sampler.
• Forgetting the audio set. On an audio-video generation, set both a VIDEO and an AUDIO parameter node and chain them. Configure only one and the other modality falls back to defaults - often the cause of "why is my audio ignoring the settings."
• Overdriving modality_scale. This is the sync strength; push it too hard and the streams over-couple (e.g. video contorting to match audio). Nudge it rather than swinging it.
• STG too high. As everywhere in LTX, heavy STG makes hair and skin look unnatural - dial stg back if faces go plasticky.