🅛🅣🅧 Multi Prompt Provider

MultiPromptProvider is the tidy way to encode a whole list of prompts in one shot. You write several prompts into a single text box, hand it a text encoder, and it gives you back all of them as a batched conditioning - no chain of individual CLIP Text Encode nodes cluttering the graph. It's a small convenience node, but when a workflow needs many prompts (a video that evolves through described beats, or any multi-segment setup), it saves a lot of wiring.

It pairs naturally with LTX's whole approach to prompting. LTX doesn't do well with short, casual text - the community consensus is you write long, detailed, per-section prompts or you get static and chaos. A node designed to accept a batch of those detailed prompts at once is the right shape for that reality.

How it works

Type your prompts into the multiline field, connect a CLIP text encoder, and the node encodes each prompt and returns the set together as one conditioning bundle. It doesn't decide when each prompt applies - that's the job of whatever consumes the output - it just does the batch encoding so the rest of your graph has clean, ready-to-use conditionings.

The inputs and outputs that matter

• prompts (STRING, multiline) - the list of prompts. This is the whole content side; enter them the way your workflow expects them delimited (typically one per line). Make each one specific - vague prompts underperform badly on LTX.
• clip (CLIP) - the text encoder that turns the prompts into conditionings. Use the one your checkpoint expects.

Output is conditionings (CONDITIONING) - the batch of encoded prompts, to route into a scheduler or sampler downstream.

How to install it

The README recommends ComfyUI Manager: Ctrl+M → Install Custom Nodes → search LTXVideo → Install → restart. Or manually:

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

then restart ComfyUI. Look under the prompt category. The node is trivial to run; the text encoder behind it is the expensive part, especially on LTX-2, whose Gemma encoder is a well-known VRAM sink. The LTX model overall wants 32GB+ VRAM and 100GB+ disk per Lightricks.

Common issues & troubleshooting

Prompts merge into one. They need to be separated in the format the node parses (usually one per line). If everything runs together, you'll get a single conditioning instead of several.

Slow or out-of-memory during encode. The text encoder is doing the heavy lifting. On LTX-2 the Gemma 3 12B encoder is huge and the usual OOM culprit - use a quantized version or the pack's cache-and-reuse text-encoding nodes to stop reloading it every run.

Which one do I use - this or LTXVMultiPromptProvider? The pack ships both, and they share the same inputs and output. From the outside they do the same job: batch prompts in, batched conditioning out. Use whichever your example workflow already references; there's no functional reason to prefer one over the other for basic use.

The prompts don't change where I expect in the video. This node only encodes. Distributing prompts across the timeline is handled downstream - verify you're feeding the output into the node that actually schedules conditioning across frames.