WaveSpeed LTX 2.3 Text-to-Video

LTX is Lightricks' video model - famous in the open-weights world for being fast and cheap to run locally. This node runs the same family, but LTX 2.3 through WaveSpeed's cloud, not on your GPU. For people who like LTX's look but don't have the VRAM (or the patience for a local LTX install), this is the compromise: you keep the model's speed and character, you rent the compute per clip.

It's part of the ERPK Collection, the community pack that wraps the WaveSpeed AI API. And there's a genuinely useful detail hidden in this node's spec that its sibling nodes don't share - more on that below.

How it works

Type a prompt, pick resolution, aspect ratio, and duration, and the node sends the job to api.wavespeed.ai, polling every 10 seconds (15-minute timeout) until the video is done. Output is a video_url STRING for the pack's Preview Anything node to preview and download. Nothing runs locally except the network request.

The differentiator: seed here is a real generation seed sent to the API. On most ERPK WaveSpeed endpoints (Kling, LTX 2 Pro), the seed widget is a cache-control dummy. Not this one - LTX 2.3's API accepts a seed, so a fixed value genuinely gives you reproducible output, not just a cached clip. That makes this the node to reach for when you need to re-run a specific look.

The inputs that matter

• prompt - the video to generate. LTX responds well to clear subject + action + camera phrasing.
• resolution - 480p, 720p, or 1080p, 720p default. Higher is slower and pricier; start at 720p.
• aspect_ratio - 16:9 or 9:16. Portrait for reels, landscape for everything else.
• duration - 5 to 20 seconds, step 1, default 5. The range is generous; LTX 2.3 is efficient enough that longer clips stay tolerable on the wallet.
• seed - the star. -1 means random; any fixed value is sent to the API for real reproducibility. Use it when you find a clip you like and want to explore variations around it.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt

Or search erpk in ComfyUI Manager → ERPK Custom Nodes. No model downloads - WaveSpeed hosts it. You need a WaveSpeed API key with credits, set via right-click canvas > ERPK Settings (Settings > ERPK > API Keys).

Common issues

• Blank video_url - the render failed upstream. Check WaveSpeed's dashboard; a rejected prompt (their content filter) is the usual cause.
• Same clip keeps coming back - you set a fixed seed and it's reproducible by design. Change it or go back to -1 for variety.
• "Prompt is required" - the node refuses empty prompts before you pay, which is the good kind of failure.
• 1080p is slow - that's the polling timeout working as intended; long generations poll for up to 15 minutes before giving up.

If you liked LTX's look but are on a laptop or a GPU with 8GB, this is the honest way to use it. And if you need audio, the separate LTX 2 Pro text-to-video node in this pack handles that - 2.3 and 2 Pro take different parameters, so the pack keeps them as distinct nodes rather than one dropdown.