Alibaba Wan 2.5 Image To Video (Official Edition)
The only way to reach it, since the weights never shipped
- image
- audio
- api_config
- video
- request_id
- urls
Here's the thing that makes this node special, and it's a bummer wrapped in a convenience: Wan 2.2 was the last open-weight Wan. Alibaba took 2.5, 2.6, and 2.7 API-only and never came back, so if you want Wan 2.5 image-to-video, a node like this is the only door - there's no local checkpoint to download, no matter how hard you search. That's the entire pitch of this one.
What this node does
BizyTRD_Wan_2_5_Official_I2V_API takes an image as the first frame and generates a video from it on Alibaba's Wan 2.5 model, running on BizyAir's servers. Flip the BizyAirPlus button ON, the node uploads your still, and the finished clip streams back into ComfyUI. Your GPU just orchestrates.
The inputs that matter
image- the first frame. The output aspect ratio tries to match it, so compose the still with the final frame in mind.resolution- the combo reads 480P by default, though the tooltip mentions Wan 2.5-I2V defaulting to 1080P. Don't overthink the mismatch: the model auto-scales to the nearest total-pixel bucket for the resolution range, so pick the ballpark and let it adjust.duration- 5 or 10 seconds. That's the whole envelope for this preview model.prompt_extend- on by default, and it's a real feature: an LLM rewrites and expands your short prompt before generation, which noticeably improves weak prompts at the cost of extra time. If you write rich prompts yourself, you can turn it off and save the wait.prompt- optional, up to 1500 characters.audio- optional custom audio: an AUDIO input (WAV/mp3, 3–30s, up to 15MB) that the model uses to drive the video, replacing auto-dubbing when present. Wan 2.5 and 2.6 support this.audio_2- on by default: auto-generate background music or SFX from the video content. Flip off for a silent take.
Shared BizyAir options: api_config (endpoint/key override) and skip_error (placeholder instead of abort).
Outputs
video- the finished clip; wire into preview or save.request_id- cloud task ID.urls- hosted result URLs.
How to install
cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt
Restart ComfyUI (or install "BizyAirPlus" via ComfyUI Manager). Deps: bizyair-cloudberry and bizytrd; no model files, because nothing runs locally. First run: account at bizyair.ai → API key → enter when the ON button prompts (or Settings > BizyAirPlus; key at ~/.BizyAirPlus/apikey.ini).
Where people get burned
Mostly the expectation that a local Wan 2.5 exists. It doesn't - and this node being cloud-only isn't a flaw, it's the model's distribution model. Also watch prompt_extend: it's free quality for short prompts but adds latency, so if your prompts are already detailed, off saves you time. And audio_2 vs audio: a custom audio input takes precedence over the auto-generated audio, so if you feed one, expect it to drive the video.
Pack note: BizyAir's malware reputation is leftover from a 2025 ComfyUI Manager auto-install bug - legitimate SiliconFlow pack, and it's removed properly with pip uninstall bizyair-cloudberry bizytrd, not by deleting the folder alone.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Input first frame picture, output video aspect ratio will try to keep consistent with the first frame picture | |
| resolution | COMBO | 480P | The model automatically scales to the nearest total pixel according to the video resolution range. WAN2.5-I2V default 1080P |
| duration | INT | 55–10 | Video length in seconds. Wan2.5-i2v-preview 5 or 10 seconds |
| prompt_extend | BOOLEAN | true | Whether prompt smart overwrite is enabled. After opening, use the large model to optimize the input prompt, improve the effect of short prompts obviously, but it will increase the time-consuming |
| audioopt | AUDIO | Customize the audio URL that the model will use to generate the video. Support for WAV/mp3,3-30 seconds, up to 15MB. It will replace auto-dubbing when available. Wan2.6/2.5 support only | |
| promptopt | STRING | The picture lives the video prompt word. Wan2.6/2.5 Max 1500 characters, wan2.2/Wanx2.1 Max 800 characters. This parameter is not valid when using video effects | |
| audio_2opt | BOOLEAN | true | Whether to automatically generate background music or sound effects. True: automatically generates audio based on video content; false: generates silent video. The audio parameter takes precedence over audio and outputs the silent video even when audio = false |
| api_configopt | BIZYAIR_OPENAPI_CONFIG | 单次请求覆盖base_url和api_key | |
| skip_erroropt | BOOLEAN | false | 开启后遇到错误不中断工作流,输出对应类型的错误占位符 |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |
| request_id | STRING | — |
| urls | STRING | — |