LTX 2.3 / videoToVideo

LTX 2.3's video-to-video node, hosted on Civitai's cloud: source footage plus a prompt becomes restyled footage, with the original clip's structure kept as conditioning. It's the gentlest edit node in the LTX 2.3 family - the guide_strength defaults to 0.2, not 1.0 - which is the whole point: this is for repainting a clip while letting the motion and composition ride, not for blowing it up and rebuilding it.

Think of it as the difference between the editVideo sibling (canny edges clamped hard at guide_strength 1.0, for keeping composition while re-rendering) and this one (whole-clip conditioning at 0.2, for style transfer and look changes that stay loose). If you want to regrade the mood, swap the season, or restyle a shot without the subject walking off-frame, videoToVideo is the node. Like every node in civitai/civitai-comfy-nodes, the job runs on Civitai's fleet, costs Buzz, and your source clip is uploaded to their servers.

How it works

Standard orchestration: videoGen workflow with the LTX 2.3 engine, submit to the Civitai Orchestration API, long-poll, download as VIDEO. The source video is conditioned on as the reference for the whole clip and guide_strength sets how strongly. The tooltip-backed detail worth internalizing: at 0.2 the model is mostly free to re-imagine look and texture but is still told where everything is; you raise guide_strength when the output drifts too far from the original action.

Inputs and outputs that matter

• prompt (required) - the restyle: "film noir", "autumn", "cyberpunk neon", etc.
• source_video (required) - the VIDEO socket with your footage.
• negative_prompt - what to avoid.
• guide_strength - 0–1, default 0.2. The creative dial: low = free restyle, high = structure locked.
• duration - tooltip: 3 or 5 seconds.
• width / height - default 1280×720; divisible by 32.
• fps - default 24; steps - default 20 (8–50); guidance_scale - default 4; generate_audio - default true.
• model - 22b-dev / 22b-distilled.
• loras - CIVITAI_LORAS socket; diffusion_model - transformer override for fine-tunes.
• quantity - 1–10 variants per job, distinct seeds.
• 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 clone into ComfyUI/custom_nodes + pip install -r civitai-comfy-nodes/requirements.txt. Only requests; no models.

Common issues

• Auth & Buzz. Civitai account with Buzz; CIVITAI_API_TOKEN for headless installs.
• "It didn't change anything." At guide_strength 0.2, subtle prompts produce subtle changes. That's the design - but if you want a stronger restyle, raise guide_strength and/or guidance_scale, don't just repeat the prompt.
• "It changed too much." If the subject starts walking off or the composition mutates, lower guide_strength back toward 0.1–0.2.
• Audio mismatch. generate_audio defaults on; if the restyled clip's sound no longer fits, flip it off and keep the source track downstream.
• Upload time. Source clips upload to Civitai; keep them short and compressed.
• Early preview. Behavior may change without notice; pin the version for anything you rely on.
• Cost. Full cloud job per run - watch the per-run cost on the node's status line.