(Deno) LTX Prompt Guide

A stock LTX text-to-video workflow makes you encode your prompt and separately attach the frame rate to the conditioning - two nodes doing what is really one job. (Deno) LTX Prompt Guide folds them together: CLIP in, positive and negative conditioning out, with the LTX frame-rate metadata written into the conditioning automatically. It also throws in a dialogue-length estimate, which is the kind of small production nicety that tells you this pack is built by someone who actually renders videos for a living.

What it does

The core is prompt encoding plus LTX's frame-rate conditioning in one node. You set frame_rate (25 default), the node adds it to the LTX conditioning and also returns it as a frame_rate output so downstream timing nodes all agree without you threading the number around manually. The negative prompt section is collapsible (show_negative_prompt), because most LTX setups run with a minimal or empty negative.

The dialogue bit is the personality: language (Auto, Korean, English, Japanese, Chinese) is used only to estimate minimum video length from quoted dialogue in your prompt - it does not translate anything. The idea is simple and practical: if your script has dialogue, you need enough frames to say it, and the estimate keeps you from generating a 4-second clip for a 10-second line. Just keep in mind it's an estimate, and it's steering you toward a minimum, not a target.

Inputs and outputs

• clip - the LTX text encoder/CLIP.
• positive_prompt - your main prompt (default text is Korean, so don't be alarmed; it's just the author's starting template).
• language - dialogue-estimate language only.
• frame_rate - written into conditioning and passed out.
• negative_prompt - editable; show_negative_prompt toggles the box.
• Outputs: positive and negative CONDITIONING (both carrying the frame-rate metadata), plus frame_rate INT for downstream nodes.

Install and context

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/Deno2026/comfyui-deno-custom-nodes.git
# restart ComfyUI

You need the official ComfyUI-LTXVideo stack and a 2.3 model (or older LTX, with matching nodes) underneath - this node is the prompt front-end of that pipeline, not a model loader. The pack's (Deno) LTX Model Loader is the natural pairing on the load side.

One honest note: LTX 2.3's prompt adherence is still its weakest axis, and the community's highest-leverage habit is seed hunting - generating several candidates on different seeds and picking - rather than polishing prompt wording. So use this node to get conditioning right and consistent, then let the seeds do the exploring. And since 2.3's rebuilt VAE broke TinyVAE previews, you'll often judge those candidates blind until decode; a real encoded preview (the pack's (Deno) Video Preview does exactly this) makes that workflow much less painful.