Alibaba Wan 2.7 Image To Video (Official Edition)

Text-to-video gives you anything, which means it often gives you nothing you wanted. Image-to-video is the opposite discipline: you hand Wan 2.7 a starting frame and it has to respect it. That's what this node does, and for the price of zero local VRAM - it's the cloud API flavor of Wan 2.7 inside ComfyUI, served by the BizyAirPlus pack. If you've generated a hero shot with Wan 2.7's image model and want it to actually move, this is the node you reach for.

It's part of SiliconFlow's BizyAirPlus pack, which is a cloud-execution layer: you edit the graph locally, flip the BizyAirPlus button ON in the action bar, and the node uploads your image and prompt to the BizyAir cloud, where Wan 2.7 does the heavy lifting.

How it works

Feed it a first_frame image and it "births" a video from that frame. Feed it both first_frame and last_frame, and it animates from one to the other - the start/end-frame trick that was a headline feature of the Wan 2.7 release. Feed it a first_clip instead (MP4/MOV, 2–10s) and it continues that video rather than starting fresh. There's even an audio input (WAV/MP3, 2–30s, ≤15MB) that drives the motion - think lip-sync and action beats - or leave it empty and the model generates its own music and sound effects.

Every run returns three outputs: video (wire it into SaveVideo), request_id, and urls for the hosted file.

The inputs that matter

The required set is small - resolution (720P default), duration, watermark, and prompt_extend - and the interesting decisions are all optional:

• first_frame / last_frame - the stills that bracket your video. Aspect ratio must sit between 1:8 and 8:1, and the tooltip is blunt that it should match the resolution you picked as closely as possible.
• first_clip - for continuation. Note the cost logic: when continuing, the output length is input length + continue length, and you're billed for the total. That's how a 15-second cap can still cost more than you expected.
• prompt - "picture student video prompt" (the author's phrasing) up to 5000 characters. It's optional because the frame alone does a lot of the work, but a good prompt is the difference between a subtle drift and an actual animation.
• duration - 2–15 seconds. Watch the continuation math above.
• prompt_extend - off by default; smart-overwrites your prompt with an LLM, better for short prompts, slower and it changes what you wrote.

Plus the usual negative_prompt, seed (-1 = server-random), and the two pack-wide safety valves: skip_error (continue with a placeholder instead of aborting the graph) and api_config (override base URL/API key for one request).

Installation

It's the same install for every node in the pack - once. ComfyUI Manager, search BizyAirPlus, install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart, then get an API key at bizyair.ai (prompt, Settings > BizyAirPlus > API Key, or the BIZYAIR_API_KEY env var). No model files download - that's the whole point.

Common issues

• Aspect ratio errors - the tooltip warns the video ratio should track your input frames. If the API rejects the request, the first thing to check is that your source image isn't more extreme than 1:8.
• Cost surprises on continuation - you're billed for the total output length, not the added seconds. Read the duration tooltip before queueing a long continue.
• Cloud run stalls or fails - key validity and network, then the ComfyUI console for the detailed error.
• Local workflows behaving weirdly - BizyAirPlus auto-installs and its action-bar toggle might be ON. Set it to OFF for local-only runs.