LTX-2.5 Fast

Here's the funny thing about Lightricks' video models: the company is the West's most committed open-weights video lab, and yet the newest generation - LTX-2.5 - is only reachable through the cloud. This node runs the "Fast" tier of that line over Runware's backend, and it earns a different kind of attention than the other cloud video nodes in this pack, because you can actually go run LTX-2.3 locally and compare. LTX's open releases have been the community's honest benchmark for "best video model you can run yourself," so when Lightricks moves the next generation behind an API, testing the hosted version against what you already have at home is the entire point.

What the LTX family built its reputation on is speed - tuned to run inside Lightricks' own consumer apps, which is why it's fast on consumer GPUs. The "Fast" tier of LTX-2.5 leans into exactly that heritage: it's the quick-iteration end of the line, for drafts and volume, with the "Pro" sibling as the quality step-up. The KB's read on the lineage is worth keeping in mind: LTX-2.3 was the release that flipped Wan users by fixing frozen image-to-video, audio artifacts, and prompt drift, and the roadmap after it promised denser models and better text encoders. LTX-2.5 is that next generation - hosted, at least for now.

The node's standout input is audio: it takes an audio clip as UUID or URL and drives the generated video from it, with the clip's length setting the video's duration. That's audio-to-video conditioning, which is rarer and more useful than it sounds - you can generate motion to match a soundtrack, dialogue, or a music bed directly instead of eyeballing sync afterward. positivePrompt is required, sentence-style. referenceImages and frameImages cover image-to-video and frame conditioning. size defaults to 1920x1080, fps is a dropdown, and settings.cameraMovement gives you a cinematic camera preset instead of leaving every shot to the model. settings.audio (note: separate from the audio input - this generates synchronized audio for the output) defaults off, which fits the fast-tier ethos. Output: VIDEO.

Install

Same pack, same one-time setup: ComfyUI Manager → search Runware → install → restart, or clone https://github.com/Runware/ComfyUI-Runware into custom_nodes and pip install -r requirements.txt. Runware API key from the dashboard (Settings → Runware API key, RUNWARE_API_KEY, or runware auth login), no model files to manage.

Should you pay for what you could run free?

That's the honest question for any Lightricks node, and the answer depends on your card and your patience. If LTX-2.3 already does your job locally, the Fast tier's only edge is convenience and the newest weights. If you're on modest hardware and want Lightricks' current quality without the quantizing and sampler-tweaking that local LTX famously requires, this is the clean door in. Watch per-run cost on the title bar either way, and remember your inputs go through Runware's cloud - for proprietary footage, local LTX-2.3 stays the privacy answer.