lightricks

Lightricks is the company behind LTX, and this node runs the Lightricks video engine on Civitai's cloud instead of your GPU. It's the hosted, zero-setup cousin of the LTX 2 / LTX 2.3 nodes elsewhere in this pack: same model family, but you never touch the weights, the Gemma text encoder, or a quantizer. Prompt in, Buzz out, video back.

There's a real pitch here if you've read the LTX story. LTX-2 (19B) brought synced audio to open video and found its killer app in lipsync and music videos; LTX-2.3 (22B) rebuilt the VAE, fixed the frozen I2V complaints, and by mid-2026 was being called the best locally-runnable video model by a meaningful share of the community. All of that quality comes with a local cost: the 22B model plus its text encoder is a serious hardware commitment. This node says "what if someone else's hardware." The catch is that the hosted Lightricks engine is a different beast from the open weights - you get settings like expand_prompt, not the distilled-LoRA wizardry people tune locally.

How it works

Standard orchestration plumbing: submit a videoGen workflow with the Lightricks engine to the Civitai Orchestration API, long-poll until done, download the MP4, return video + workflow_id + raw_json. Your prompt and any source image go to Civitai's servers. The one engine-specific toggle worth understanding is expand_prompt (default on) - Lightricks' hosted pipeline runs your prompt through an expansion step before generation, which helps weaker prompts but can also add stuff you didn't ask for. Flip it off if the output keeps drifting from your intent.

Inputs and outputs that matter

• prompt (required) - the clip.
• cfg_scale (required) - default 4, range 0–100.
• negative_prompt - what to avoid.
• frame_rate - default 25.
• duration - 1–30 seconds, default 5.
• seed - 0 for random.
• steps - default 20, range 10–50.
• aspect_ratio - 1:1, 16:9, 9:16, 3:2, 2:3.
• source_image - optional image-to-video start frame (IMAGE socket; accepts URL/DataURL/Base64).
• expand_prompt - default true; the prompt-expansion toggle described above.
• api_config - auth override.

Outputs: video, workflow_id, raw_json.

Installing it

Same pack as all Civitai nodes. 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

Just 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.
• expand_prompt drift. If generations keep adding elements you didn't write, try expand_prompt = false and write your own richer prompt. The LTX prompting lore applies here too: long, detailed, camera-aware prompts beat short ones.
• Hosted ≠ local tuning. The community's distilled-vs-dev advice (dev = more steps, better prompt adherence; distilled = fast, 8 steps) is about the local checkpoints and doesn't map 1:1 onto Civitai's hosted engine. Don't over-fit your settings to local recipes.
• Early preview. The pack is under active development; behavior may change without notice.
• Cost. Video jobs bill per run; longer durations cost more. Read the per-run cost on the node's status line.