Wan Image to Video

If you've been generating Wan 2.2 locally - and a huge part of this community has - you already know what Wan image-to-video feels like. This node is the next two generations of that, served over the API. It takes a first-frame image and a prompt, and animates the frame forward with the Wan 2.5 or Wan 2.6 model, with a couple of extras the local models never got: native audio and a 15-second option.

It's an API node, same as every other entry in ComfyUI's built-in partner/video category. It ships with ComfyUI core, uploads your image to Comfy Org's backend, calls Alibaba's Wan API on your behalf, and bills your Comfy account in credits. There are no Wan 2.5/2.6 weights to download - those stopped shipping after 2.2 - so this is what "newer Wan" means now. Your GPU is a spectator.

The inputs that matter

• model - wan2.6-i2v (default) or the older wan2.5-i2v-preview. 2.6 is the one that unlocks the long duration and better quality.
• image - the first frame. The whole clip starts here.
• prompt - describes the elements and visual features, and it natively supports English and Chinese. Useful if you work bilingually; Wan understands both.
• duration - 5, 10, or 15 seconds. The catch, spelled out in the tooltip: 15 seconds is only available on the Wan 2.6 model. On 2.5, 15 fails validation.
• audio - an optional audio input that must be "a clear, loud voice, without extraneous noise or background music." Feed a voice track here and the video generation can sync to it. generate_audio (off by default) instead has the model produce audio automatically when no input is provided.
• prompt_extend (default on) - server-side prompt enhancement. shot_type (single or multi) only takes effect when prompt_extend is on, and controls whether the output is one continuous shot or multiple shots with cuts. Turn prompt_extend off if you've written the prompt exactly as you want it.

Also optional: negative_prompt, resolution (480P/720P/1080P), seed, watermark.

Output is a single VIDEO.

The honest take

Wan's hosted line got a lukewarm community reception - the same crowd that called 2.7 "shit" mostly shrugged at 2.5 and 2.6. But image-to-video is where Wan has always been strongest, and the first-frame anchoring fixes the consistency problems that plague pure text-to-video. If you already have a Wan workflow, the I2V muscle memory carries over: describe what's on the image, then describe the motion.

Where people get burned: the 15-second duration on the wrong model, the audio input that isn't a clean voice track (garbage in, garbage sync out), and the general expectation that because it's named "Wan" it's free. It isn't - it's per-second credits. Treat this as the convenient hosted upgrade to the model you already love, not a replacement for the open one.