wan / v2.2 / fal / text-to-video

Wan 2.2, text-to-video, running on fal's hosted fleet via Civitai's Orchestration API. No local GPU, no model files, no download - you prompt, Civitai's cloud (through fal) renders it, and the clip comes back as a native VIDEO socket in your ComfyUI graph. Billed in Buzz, like every node in the official civitai-comfy-nodes pack.

This is the node to reach for when you want the open-model quality that made 2.2 the community's video default - expressive motion, better facial emotion than 2.1 - without owning the hardware to run it. The catch is the same one as running it locally: 2.2 is slow and wants good prompts, and here "slow" is also "metered."

The inputs that matter

prompt and cfg_scale (default 4) are required. Then:

• resolution - 480p or 720p (default 720p).
• aspect_ratio - 1:1, 16:9, 9:16, 4:3, 3:4, 4:5, 5:4 (default 16:9).
• shift - 1.0 to 10.0, default 5.0. Wan's training-shift dial; leave it alone unless you're tuning deliberately.
• use_turbo - off by default. Faster/cheaper, visibly flatter motion. Draft with it.
• interpolator_model - film or rife (default film), for frame interpolation.
• negative_prompt - steering, useful here.
• enable_prompt_expansion - off by default; fal's prompt rewrite.
• enable_safety_checker - off by default.
• loras - CIVITAI_LORAS socket from the Civitai LoRA Selector; trigger words go in prompt.
• source_image exists too, but for text-to-video leave it empty - or, honestly, if you have a frame you like, switch to the i2v sibling and get better consistency.
• duration 1–30s (default 5), frame_rate 24, steps 10–50, seed, api_config.

Outputs

Standard trio: video (VIDEO) into a Save Video / preview node, workflow_id (STRING) for the Civitai sidebar, raw_json (STRING) for response and cost.

Install and auth

One-time, pack-wide. ComfyUI Manager → search Civitai Comfy Nodes → install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # just `requests`

Auth: Civitai Auth node, CIVITAI_API_TOKEN (headless), or sidebar sign-in.

Troubleshooting and takes

• The prompt is everything here. No anchor frame means 2.2 has to invent the scene. Write motion and camera explicitly - "slow dolly in," "subject turns to camera" - or you'll get a pretty but static clip that technically moves.
• Slow is baked in. 2.2's quality costs time; locally that was ~25 minutes a clip on a 5090 without speed-ups. Here it's queue time plus Buzz. The pack launched June 2026 during Civitai's GPU crunch, so long waits and timeouts were the top real complaints. Default timeout 30 min - raise via Civitai Auth timeout_minutes or CIVITAI_COMFY_TIMEOUT.
• Turbo changes the character. use_turbo is a real quality trade, not a free lunch. The community's standing verdict: speed LoRAs "destroy everything that is good" in 2.2's motion and skin. Fal's turbo is gentler than the speed-LoRA disaster, but the same idea applies.
• 2.2 vs 2.1. 2.1 (also in this pack via fal) is simpler and battle-tested; 2.2 is the better-looking, more expensive one. Neither is wrong.
• Early preview. Generated nodes "may change without notice."

Reach for this when you want open-model Wan 2.2 text-to-video with a modest meter and no hardware. Have a frame? The image-to-video sibling is where Wan really shows off.