Alibaba WAN 2.7 Image-to-Video

WAN 2.7 Image-to-Video is Alibaba's Wan video model, version 2.7, served through WaveSpeed as a hosted API. The context you need before using it: Wan is the family that put open video on the map at 2.1 and 2.2, then went API-only - the KB's Wan panel is unambiguous that 2.5, 2.6, and 2.7 have no weights. So this node is not a shortcut to running a local model; it's a paid API wrapper around a model that was never released open. The node itself is the practical way to get at it.

What you get for the money is a current commercial Wan: an image in, a 2–15 second clip out, with an optional end frame (last_image) for smooth transitions, a negative prompt, optional audio guidance, and prompt expansion. That's a real feature set, and the end-frame continuation is the standout - it's the same idea as the Seedance I2V node's end frame: pin the destination and let the model bridge the motion.

The inputs

• image - the source image URL (first frame of the video)
• prompt - what happens in the clip
• last_image - optional end-frame image URL for a smooth transition
• negative_prompt - elements to exclude
• audio - optional audio URL to guide generation
• duration - 2 to 15 seconds, default 5
• resolution - output resolution dropdown
• enable_prompt_expansion - automatically enrich the prompt before generation
• seed - reproducibility, -1 for random
• client - optional WaveSpeed client, falls through to ERPK Settings

The output is video_url (STRING) → Preview Anything. Note the image inputs here are URL strings, not IMAGE tensors - unlike the Seedance I2V node, there's no tensor input, so you'll need your frame hosted somewhere or passed as a path. If you're generating the frame in the same workflow, that's an extra step this node doesn't save you.

Install and setup

cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk && pip install -r requirements.txt

Or "ERPK Custom Nodes" from ComfyUI Manager, restart, and set a WaveSpeed API key in ERPK Settings. Generations poll every 10 seconds with a 15-minute timeout.

Troubleshooting

The empty-input errors are your friends here: "Image URL is required" stops you before you spend money on a frameless generation. enable_prompt_expansion defaults to off - if your clips come back underdeveloped, that toggle is the free lever to pull before you start rewriting prompts. And mind the duration range: this node goes 2–15 seconds, which is shorter at the top end than Seedance's 4–15. For the Wan-vs-Seedance choice: Wan is generally the sturdier all-rounder for image-to-video continuity, Seedance leans into reference-based style control and native audio. Both are billed through the same WaveSpeed key, so try a short clip on each and let your use case decide.