LTX 2 / createVideo

This node runs LTX 2 - Lightricks' 19B model, the one that brought synchronized audio and video into a single open generator - on Civitai's cloud, from inside ComfyUI. It does text-to-video and optional image-to-video (via the images input), billed in Buzz, with no weights on your disk.

Understanding where LTX 2 sits matters before you spend on it. Released January 2026, LTX 2 was the model whose synced audio and voice-carrying continuation wowed people, but whose I2V froze, whose audio artifacted, and whose 22GB Gemma 3 text encoder caused most of the community's OOMs (the KB's ltx-video doc has the whole saga, including the Kijai one-liner fix). Then LTX 2.3 shipped in March 2026 and fixed the frozen I2V, rebuilt the VAE, and added native portrait. This pack has separate LTX 2.3 nodes for that. So the honest read: pick LTX 2 when you specifically want the 19B behavior or its settings; pick the LTX 2.3 nodes when you want the model Civitai is more likely to keep tuned. The older node also lacks the 2.3 nodes' diffusion_model override and quantity fields - a simpler surface, for better or worse.

How it works

Standard orchestration plumbing: assemble a videoGen workflow with the LTX 2 engine, submit to the Civitai Orchestration API, long-poll until the job succeeds, download the MP4 as a native VIDEO with workflow_id and raw_json alongside. The hosted engine runs a Comfy workflow server-side; you're renting both the GPU and the model.

Inputs and outputs that matter

• prompt (required) - the clip. LTX rewards long, camera-aware prompts; the 2.x era was brutal on short ones.
• model - 19b-dev (default) or 19b-distilled. Distilled is the 8-step fast path (the community standard for quick drafts); dev wants ~20 steps and follows prompts better.
• duration - tooltip: 3 or 5 seconds.
• width / height - default 1280×720. LTX-2 renders internally at half your requested resolution and upscales, which is why people say matching Wan's 720p quality took a 1080p request.
• fps - default 24. For I2V, the community found 48fps helps the temporal compression glitches.
• generate_audio - default true. The synced-audio feature; flip off for silent drafts to save cost.
• guidance_scale - default 4; steps - default 20 (8–50).
• negative_prompt - what to avoid.
• loras - CIVITAI_LORAS socket; wire a Civitai LoRA Selector to stack LoRAs.
• images - optional IMAGE socket for image-to-video.
• api_config - auth override.

Outputs: video, workflow_id, raw_json.

Installing it

ComfyUI Manager → Custom Nodes Manager → "Civitai Comfy Nodes" (publisher civitai) → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Only requests; no model or text-encoder downloads.

Common issues

• Auth & Buzz. Civitai account with Buzz; CIVITAI_API_TOKEN for headless installs, Auth node or sidebar otherwise.
• I2V can be static - it's the model, not you. Frozen or slow-pan I2V was the #1 LTX-2 complaint; 2.3 fixed it. If animated-still outputs frustrate you, try 48fps and a motion-heavy prompt, or just move to the LTX 2.3 nodes.
• Distilled vs dev. Don't run dev at 8 steps - it needs ~20. Distilled at 8 is the fast draft recipe.
• Early preview. The pack is under active development; behavior may change without notice.
• Cost. Every run bills Buzz; longer durations and audio cost more. Read the per-run cost on the node's status line.