LTX 2.5 / videoToVideo

Video-to-video is where you feed an existing clip and get a re-imagined version: same motion, new look. In the LTX 2.5 family this node is the light-touch restyle tool, and the one input that defines it is guide_strength - which defaults to a gentle 0.2 here. That low default is the tell: LTX 2.5's video-to-video is meant to reinterpret your source while keeping its structure, not to overpower it. Crank it up and the model takes over; keep it low and it's more of a re-grade.

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

Inputs that matter

• source_video (required) - a VIDEO socket; wire your clip in.
• prompt (required) and negative_prompt - describe the target look: "converted to watercolor", "night scene, neon".
• guide_strength - 0.0 to 1.0, default 0.2. The creative dial. Low = faithful to source, high = wilder reinterpretation. This is the input to experiment with first.
• duration (3–20s, default 5), width/height (1280×720), fps (24), seed, guidance_scale (4), steps (20) - the usual LTX surface.
• generate_audio - on by default (LTX's native synchronized audio).
• model (22b-dev), loras (wire a Civitai LoRA Selector), diffusion_model (point at a community fine-tune like SulphurAI/Sulphur-2-base), quantity (N billed renders with distinct seeds).

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

guide_strength is the trap dial. People wire in a clip, leave the default at 0.2, and get "nothing changed" - that's not a bug, that's the setting working as designed. It's a subtle-to-loud scale; for actual restyling you'll usually want it well above the default. Upload cost is real. Your source clip ships to Civitai's cloud and every second of output is billed; trim the source to what you actually want restyled. Your clip leaves your machine, full stop. And the pack is early preview - behavior can change without notice. If your aim is re-editing rather than restyling, the sibling editVideo node (default guide strength 1.0) is the more aggressive tool; pick by how much you want the source to survive.