Wan 2.6 - Image to Video

I2V is where most people actually live, because anchoring frame one does more for consistency than any sampler trick. Wan 2.6 is the December 2025 closed-weight release - no local weights, API only, and the community was blunt about it. What the API does give you is fifteen-second clips at 1080p with optional sound, from a single image, without touching your VRAM budget.

This is the mainstream I2V node in the pack's Wan folder: image in, prompt in, clip out. Reach for it when you've already got the frame right and just need it moving.

How it works

Required: prompt and image_urls. The prompt field accepts Chinese and English, minimum 2 characters, max 5000. image_urls is a plural-named IMAGE socket, so a batch is accepted - practically, you feed it the frame you want animated.

The node uploads your image to KIE, POSTs to /api/v1/jobs/createTask with marketplace model wan/2-6-image-to-video, polls /api/v1/jobs/recordInfo until the task resolves, then downloads the finished MP4 into ComfyUI/output/KIE-Results/videos and hands back a native ComfyUI VIDEO. Your ComfyUI only ever sees the HTTP client: requests, Pillow, numpy, PyYAML.

Inputs that matter

• image_urls - a clean first frame. Resolution here affects the model's framing more than your output tier does.
• prompt - motion and camera language, not a description of the picture. "Anthropomorphic fox singing a Christmas song at the rubbish dump in the rain" is the shipped default, which tells you the model wants a whole scene beat rather than a caption.
• duration - 5, 10, or 15 seconds. The long end is where this node beats the 2.5 generation.
• resolution - 1080p (default) or 720p. The single biggest cost lever.
• multi_shots - off by default. On, it builds the clip as multiple shots with transitions instead of one continuous take.

Then the pack's standard trio: nsfw_checker off, timeout_seconds at 1200, callback_url empty.

Outputs

video to a save/preview/combiner. url and all_urls_json if you want the hosted file or every artifact the endpoint returned; task_id to trace a job; raw_json for the full provider response; credits_consumed and credits_left for the money, with a receipt printed on the node after each run. Fifteen-second 1080p clips are exactly the thing that receipt exists for.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
cd ComfyUI-KIE-Nodes-Next && pip install -r requirements.txt

Restart ComfyUI and look in KIE Next / Video / Wan. The folder tree is generated from KIE's live API catalog on startup - the pack intentionally doesn't present one generic "any API" node, so each model gets its own class and its own inputs.

One-time key: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. Validated before it's saved; masked fingerprint shown; full key stays backend-side and out of the workflow JSON. KIE_API_KEY for headless installs. KIE bills credits per second of video, tiered by resolution, and there's no subscription attached.

Common issues

A 15-second 1080p run costs multiples of a 5-second 720p one. That's linear metering on duration and resolution at once. Storyboard at 720p/5s, then spend on the takes that earn it.

The clip comes back as separate shots. multi_shots got turned on.

Timed out at 20 minutes. Raise timeout_seconds (default 1200) for long high-res jobs during peak hours. Cancelling doesn't refund the generation.

Nothing renders in the UI. Not an output node - attach a video save or preview.

Errors referencing comfy_api.latest.InputImpl.VideoFromFile. Your ComfyUI predates the native VIDEO type. Update it.