LTX 2.5 / extendVideo

Every video model has a ceiling on how long a single clip can be. LTX 2.5's answer is continuation: this node takes your existing clip and keeps generating past its end - and its distinguishing input is num_frames, default 121, which is how many frames of new footage get added. That's the whole game here. Want a longer continuous take without cutting to a new generation? This is the node. The prompt tells the model where the story goes next; the source video tells it how we got there.

It's part of civitai-comfy-nodes, Civitai's official ~160-node pack generated from its Orchestration API spec. Cloud job: LTX 2.5 runs through Comfy workflows on Civitai's fleet, you get the full parameter surface, and everything bills in Buzz. Submit, long-poll, get a native VIDEO tensor.

Inputs that matter

• source_video (required) - the clip to continue from. Wire it in as the native VIDEO type.
• prompt (required) and negative_prompt - where the continuation goes: "she turns toward the camera and speaks", "the camera pulls back to reveal the city".
• num_frames - 121 frames of extension by default. At the default 24 fps that's about 5 seconds added. This is the cost lever: every added frame is billed.
• duration (3–20s), width/height (1280×720), fps (24), seed, guidance_scale (4), steps (20), generate_audio (on) - the usual LTX 2.5 surface.
• model (22b-dev), loras, diffusion_model (fine-tune override), quantity (N billed renders).

Outputs: video (VIDEO), draft_cache_audio (AUDIO), draft_cache_video (VIDEO), workflow_id and raw_json (STRING).

Install and auth

Standard for the pack - official, Comfy Registry:

# ComfyUI Manager: search "Civitai Comfy Nodes"
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Restart, authenticate via a Civitai Auth node, CIVITAI_API_TOKEN, or sidebar OAuth.

Common issues

num_frames vs duration is the confusion everyone hits. On the other LTX 2.5 nodes, duration is the whole job; here duration and num_frames both exist and the extension is governed by num_frames. If your clip doesn't get longer, check you moved num_frames, not just duration. Longer isn't linear-expensive, it's more-expensive - each extension frame bills, and chaining two extends bills twice; plan the shot before you stack continuations. Continuity drift is a real risk - at 121 frames the model can wander from the source's style; strong prompts and matching settings reduce it but don't kill it. Your clip leaves your machine to a third-party server. And the pack is early preview - behavior can change without notice.