☁️BizyAir Wan_T2V_Pipeline

Read the output type before you build around this one: it returns video_url, a plain STRING, not a VIDEO object. Every other Wan and Veo and Vidu node in this pack hands you back a proper VIDEO you can wire straight into a preview or save node. This one alone gives you a URL string instead - which means it won't drop into a Video Preview node the way you'd expect, and you'll need a way to actually fetch or display that URL yourself. If a generation "succeeded" with no visible output, this is almost certainly why.

That quirk aside, this is BizyAir's cloud-hosted text-to-video path for Wan's older, non-diffusers pipeline - paired with BizyAir Wan_Model_Loader upstream, which currently offers the 1.3B checkpoint.

What it does

Takes a loaded Wan model, a prompt, and generation settings, and produces a video on BizyAir's cloud - no image input at all, pure text-to-video. The devs left the prompt field's default value as a small, cute tell of the pack's personality: "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage." You'll obviously want to replace it, but it's a nice touch as far as placeholder text goes.

Inputs and outputs that matter

• model - from BizyAir Wan_Model_Loader, required.
• prompt - the scene description, multiline.
• resolution - five presets: 480*832, 832*480, 624*624, 704*544, 544*704. A square option plus both portrait and landscape variants, all in Wan's standard 480p-class resolution range.
• sampling_steps (default 50, capped at 50) - notice the default is the ceiling here. There's no headroom to push steps higher for extra quality on this node the way other Wan pipelines in the pack allow; you can only go down from 50, not up.
• guidance_scale (default 6, range 0–20) - this is CFG. Wan generally wants something in the 3.5-and-up range for decent prompt adherence; the default here is a reasonable, safe starting point.
• shift_scale (default 8, range 0–20) - this is the timestep-shift knob, a genuinely model- and configuration-specific control on flow-matching-family models, distinct from CFG and distinct from the scheduler itself. It's the parameter that effectively replaced a lot of what scheduler tuning used to do on older DDPM-style models. Different resolutions and prompt content can want meaningfully different shift values - don't treat the default as universally correct, it's worth actually experimenting with if results feel off in a way steps and CFG aren't fixing.
• seed (default -1) - -1 means random; set an explicit number to reproduce a result.
• negative_prompt (default "Low quality, blurry") - a sensible generic default, already filled in for you.

Output is video_url, a STRING. Not a VIDEO. Worth saying twice given how easy it is to assume otherwise coming from any other node in this pack.

How to install it

• ComfyUI Manager - search "BizyAir", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git, restart.
• Comfy CLI - comfy node install bizyair.

Set your API key first - the in-app login prompt or bizyair.siliconflow.cn. This node runs against BizyAir's cloud and spends credits per call.

Common issues & troubleshooting

Nothing shows up in your Video Preview node. Because there's nothing to preview directly - the output is a URL string, not a video object. Wire it into a text-display node to actually see it, or handle fetching the video yourself downstream.

Results feel off in a way steps/CFG tweaks don't fix. Try shift_scale before assuming the model itself is the problem. It's a real, meaningful lever on flow-matching models and it's specific enough to your resolution and prompt that the default won't always be right.

Sampling quality feels capped. sampling_steps tops out at 50 with the default already sitting there - there's no "just add more steps" option on this particular node.