wan / v2.2-5b / fal / image-to-video

Wan 2.2's 5B dense model was the one that promised 8GB-VRAM video generation locally - and this node hands you that same model without any VRAM at all, running on fal's fleet through Civitai's cloud. Image in, clip out as a native VIDEO socket, billed in Buzz. It's part of the official civitai-comfy-nodes pack.

The image-to-video variant is where Wan earns its reputation. Anchor a first frame and 2.2 animates it with genuinely expressive motion - this is the "more motion, more emotion" release that dethroned 2.1 for quality-first work. And because it's the 5B, you're getting the lighter, faster sibling of the 27B MoE, which shows up as shorter queues and smaller bills.

The inputs that matter

prompt and cfg_scale (default 4) are required. Then the stack, which is the richest in the Wan family:

• source_image / images - wire your starting frame in. This is the whole point of the i2v variant.
• resolution - 480p, 580p, or 720p (default 720p). The 5B's practical ceiling; don't expect 1080p here.
• aspect_ratio - 1:1, 16:9, 9:16, or auto (default 16:9). auto matches your source image.
• use_distill - the distilled/fast inference toggle, off by default. This is fal's speed-model option.
• use_fast_wan - the "fast Wan" toggle, off by default. Different fast path; both trade a little quality for turnaround. Worth knowing: the community's verdict on Wan 2.2 speedups is mixed - they accelerate but flatten the motion and detail that make 2.2 good.
• num_inference_steps - 2 to 50, default 20. Lower with a distill/fast model.
• interpolator_model - none, film, or rife (default film). Frame interpolation between generated frames for smoother motion.
• negative_prompt - steer away from artifacts.
• enable_prompt_expansion - off by default; fal's prompt rewrite.
• enable_safety_checker - off by default. On, and fal's safety pass runs before output.
• loras - CIVITAI_LORAS socket from the Civitai LoRA Selector. Trigger words go in the prompt; the selector's keywords field is only a reminder.
• duration 1–30s (default 5), frame_rate (default 24), steps (10–50), seed, api_config.

Outputs

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

Install and auth

One-time pack setup. 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 text-to-video sibling has the same inputs. This node's twin - CivitaiVideoGenWanV225bFalTextToVideo - is identical except it starts from prompt alone. Pick this one when you have a frame.
• Fast modes change the look. use_distill/use_fast_wan are how fal keeps costs down, and they visibly soften output. Draft with them, run finals without.
• Interpolation surprises. With film/rife on, output is smoother but not always more natural - interpolation artifacts are a thing. Compare none once.
• Slow / timeout. Pack launched June 2026 during Civitai's GPU crunch; long queues and timeouts were the top complaints. Default timeout 30 min; raise via Civitai Auth timeout_minutes or CIVITAI_COMFY_TIMEOUT.
• Early preview. Generated nodes "may change without notice."

Reach for this when you want Wan 2.2 quality on a first frame without a GPU, and you're happy living at 720p - the 5B's home turf.