LTX 2.5 / editVideo

The LTX 2.5 family has three video-in video-out nodes, and they look near-identical until you check the one number that matters: the default guide strength. videoToVideo sits at 0.2 (gentle restyle), editVideo at 1.0 - full grip. Same core surface, opposite intent. If you want the model to actually take your source clip and remake it - new action, new object, a genuinely different scene that still reads as "that clip" - this is the node. It's the aggressive sibling.

Like everything in civitai-comfy-nodes - Civitai's official ~160-node pack generated from its Orchestration API spec - it's a cloud job. LTX 2.5 runs through real Comfy workflows on Civitai's fleet, so you get the full diffusion parameter set with the GPU rented. Submit, long-poll, get a native VIDEO tensor. Billed in Buzz.

Inputs that matter

• source_video (required) - VIDEO socket for your clip.
• prompt (required) and negative_prompt - the edit instruction. At guide strength 1.0, the prompt has real authority over what the output becomes.
• guide_strength - default 1.0, range up to 1.0. Back it off toward 0.5 for a lighter edit, keep it near max for a full re-imagining.
• duration (3–20s, default 5), width/height (1280×720), fps (24), seed, guidance_scale (4), steps (20), generate_audio (on) - the standard LTX 2.5 surface.
• model (22b-dev), loras (Civitai LoRA Selector), diffusion_model (community 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 pack install - 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

Picking the wrong sibling is the classic mistake. videoToVideo, editVideo and extendVideo all take a source_video and a prompt; only the guide strength (and num_frames on extend) really separates them. If your "edit" comes out barely changed, you're almost certainly on videoToVideo at 0.2. At guide 1.0, prompt quality is everything - a vague instruction produces a clip that's changed but meaningless. Cost and upload apply as everywhere: source ships up, output is billed per second, so trim first. Your clip leaves your machine. And early preview: the pack's nodes and behavior can change without notice. One more: LTX's audio remains the slightly fragile part of the stack - if the re-cut comes back with muddled sound, that's a known family weakness, not your workflow.