Nodes/ComfyUI MediaHub/Wan Bailian CN T2V Config
ComfyUI Node

Wan Bailian CN T2V Config

Wan 2.7 video from a text prompt, with zero local VRAM

By vantang·Created 3 months ago·Updated 3 months ago· 3
Wan Bailian CN T2V Config
    • request
    prompt
    model
    resolution
    ratio
    duration5
    watermarkfalse
    seed-1
    prompt_extendtrue
    negative_prompt
    audio_url

    If you want Wan 2.7 video, you can't run it - there are no weights. Wan 2.2 was the last open Wan; 2.5, 2.6, and 2.7 all shipped API-only, and the "download Wan 2.7 locally" pages cluttering search results are fake. So the real question isn't "what GPU do I need," it's "how do I reach the API." Wan Bailian CN T2V Config is the ComfyUI answer: text-to-video on Alibaba's closed wan2.7-t2v model, billed by the second, with no local rendering hardware and no model files.

    It's part of ComfyUI MediaHub (vantang), the pack that routes a handful of cloud video APIs - MiniMax, Agnes, Seed Volcengine, and Alibaba's Bailian/DashScope platform - through ComfyUI's normal node graph. Like its image sibling, this is a config node: it builds a request, and the pack's generic execution nodes do the talking.

    The workflow

    Video is async, so the chain has two extra hops:

    Prompt -> WanBailianCNT2VConfig -> MediaHub Submit Video -> MediaHub Resolve Video -> SaveVideo
    API Key ----------------------------> MediaHub Submit Video      ^
    API Key ---------------------------------------------------------+
    
    • MediaHub Submit Video POSTs your request to DashScope's video-synthesis endpoint with the X-DashScope-Async header, gets back a task_id, and wraps it in a job.
    • MediaHub Resolve Video polls the task endpoint every 5 seconds (up to 10 minutes), downloads the finished MP4 to a temp folder, and hands you a ComfyUI VIDEO output that plugs straight into SaveVideo.

    The config node's only output is request (typed MEDIAHUB_VIDEO_REQUEST). The key, the polling, the download - all of that belongs to the execution nodes.

    The inputs that actually matter

    • model - a single choice, wan2.7-t2v. No picking.
    • resolution - 720P or 1080P. 1080P is slower and pricier; start at 720P.
    • ratio - 16:9, 9:16, 1:1, 4:3, or 3:4.
    • duration - seconds, default 5. The API accepts 2–15, and the special values -1 (let the API decide) and 0 (API default) also pass validation.
    • prompt_extend - on by default. The API rewrites and expands your prompt before generating. Leave it on until you have a reason to fight it.
    • negative_prompt - an actual negative prompt field, rarer than it should be on hosted APIs. Only sent to the API if non-empty.
    • audio_url - a URL to an audio file if you want audio-driven generation; leave it empty otherwise. The API also does auto sound generation.
    • seed - -1 means random (and isn't sent); set a value for repeatable takes.

    Installing it

    Same as the rest of the pack. In ComfyUI Manager search "MediaHub" (registry id mediahub-cloud), or the manual route:

    cd ComfyUI/custom_nodes
    git clone https://github.com/vantang/ComfyUI-MediaHub.git
    /path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt
    

    Dependencies are just requests, Pillow, and numpy - nothing heavy, no model downloads. What you actually need is a DASHSCOPE_API_KEY from Alibaba's Bailian console.

    Where people get burned

    • The same regional wall as the image node. Wan 2.7 on DashScope is a China-console offering (bailian.console.aliyun.com, Beijing endpoint), and international Alibaba accounts don't carry it. Outside mainland China you'll likely fight account verification to get a working key - the most common reason this graph dies with an auth error before it ever produces a frame.
    • The queue blocks while it polls. Resolve Video sits in your graph for however long the job takes, up to the 600-second timeout, after which it errors out. At 1080P that's a real wait; plan around it instead of assuming your graph hung.
    • Always-run, and video is billed per second. Both execution nodes are marked always-changed, so they fire on every queue with no caching. A 15-second 1080P clip isn't a throwaway test - be deliberate with it.
    • Key hygiene. The key widget is password-masked, but ComfyUI can still serialize the value into workflow JSON and file metadata. Strip it before sharing and rotate if it leaks; the README is blunt about this.
    • Manage expectations. Wan 2.7's reception among the local crowd has been mixed - takes range from "fine for static and slow-pan work" to "worse than 2.2." Some of that is sour grapes from people who can't run it at all. As an API it's legitimately usable; just don't believe the hype in either direction.

    If you have DashScope access and you're already paying for cloud video, this gets closed Wan 2.7 into ComfyUI with about five nodes and no rendering hardware. That's a genuinely rare thing for a ComfyUI custom node to offer.

    CategoryMediaHub/Provider/Wan Bailian CN/Video

    Inputs (10)

    NameTypeDefaultDescription
    promptSTRING
    modelCOMBO1 options: wan2.7-t2v
    resolutionCOMBO2 options: 720P, 1080P
    ratioCOMBO5 options: 16:9, 9:16, 1:1, 4:3, 3:4
    durationINT5-1–15
    watermarkBOOLEANfalse
    seedINT-1-1–2147483647
    prompt_extendBOOLEANtrue
    negative_promptSTRING
    audio_urlSTRING

    Outputs (1)

    NameTypeDescription
    requestMEDIAHUB_VIDEO_REQUEST