🅛🅣🅧 LTXV Multi Prompt Provider

When your LTX video needs to change over its length - a shot that starts calm and turns chaotic, a sequence that moves through several described beats - you need more than one prompt. LTXVMultiPromptProvider takes a batch of prompts in one text box, encodes them all through your text encoder in a single node, and outputs a set of conditionings you can drive across the clip. Instead of stacking a dozen CLIP Text Encode nodes and wiring each by hand, you write your prompts as a list and get them back ready to use.

This fits LTX's temperament well. LTX is famously prompt-hungry - short prompts give you static or chaotic garbage, and the community's standing advice is long, specific, per-section prose (people literally "write almost a paragraph for each short section of the script"). A node built for feeding multiple detailed prompts at once is exactly the ergonomics that workflow wants.

How it works

You put multiple prompts into the multiline box and connect a CLIP (the text encoder). The node runs each prompt through the encoder and returns them together as a batched conditioning. Downstream, those conditionings get applied across your video - the specifics of how they're distributed depend on the workflow consuming them, but the node's job is the batch encode: many prompts in, one tidy conditioning bundle out.

The inputs and outputs that matter

• prompts (STRING, multiline) - your prompts, entered as a list in the box. This is where the multi-prompt content lives. Keep each one detailed; LTX rewards specificity and punishes terseness.
• clip (CLIP) - the text encoder to run them through. For LTX-2 that's the Gemma-based encoder; wire in whatever CLIP/text-encoder your checkpoint uses.

Output is conditionings (CONDITIONING) - the batch of encoded prompts, to feed into the node or sampler that schedules them.

How to install it

Install the pack from ComfyUI Manager (Ctrl+M → Install Custom Nodes → search LTXVideo → Install → restart), the README's recommended path. Or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/Lightricks/ComfyUI-LTXVideo

then restart ComfyUI. It's in the prompt category. The node is light, but the text encoder it drives is not - LTX-2's Gemma 3 12B encoder is the model's biggest VRAM hog, and the checkpoint overall wants 32GB+ VRAM and 100GB+ disk per Lightricks.

Common issues & troubleshooting

Only one prompt seems to register. Check how you're separating them in the box - the node expects a list, so each prompt needs to be delimited the way the workflow expects (typically one per line). If they're mashed into one blob, you get one conditioning.

Encoding is painfully slow or OOMs. That's the Gemma text encoder, not this node. It's over 22GB in fp16 and is the classic LTX-2 memory sink. Use a quantized Gemma (community fp8/4-bit repacks) or the pack's text-encoding nodes that save and reuse encodings so you're not reloading the encoder constantly. --reserve-vram helps on tight cards.

The prompt changes don't land where you want in the clip. This node encodes; it doesn't schedule. How the batched conditionings map onto the timeline is up to the downstream node. Make sure you're feeding them into whatever your workflow uses to distribute prompts across frames, and that the count lines up.

Weak or ignored prompt following in general. LTX's prompt adherence is its weakest axis, full stop. Long, concrete prompts help; so does seed-hunting (generating several candidates and picking) rather than endlessly reworking wording. This node makes multi-prompt setups tidy - it can't make LTX obey a vague prompt.

Note the pack also ships a plain MultiPromptProvider with the same inputs and output. If you see both, they behave the same way.