☁️BizyAir Hailuo2.3 Text To Video

What it is

Hailuo is MiniMax's video model, and like the rest of the closed Chinese video lineup this pack wraps, it's never released weights and never will. The KB's summary of the field lumps it in plainly: "Vidu, Kling, Hailuo, all closed source." This node turns a text prompt straight into a video clip via Hailuo's API, with BizyAir handling the account and billing plumbing on your behalf.

It's the text-to-video counterpart to the pack's Kling and Sora nodes - no starting image required, just a prompt. Reach for it when you want a from-scratch clip rather than animating an existing still.

How it works

You write a prompt describing the scene and motion, pick a duration and resolution, and Hailuo's servers generate the clip. There's no local computation at all; ComfyUI submits the job and waits.

Inputs and outputs that matter

prompt is the only creative input - there's no negative prompt or reference image on this node, so everything you want in the clip has to go in one description. model is currently fixed to MiniMax-Hailuo-2.3 (a single-choice enum). duration is either 6 or 10 seconds. resolution is 768P or 1080P, with 1080P as the default - worth dropping to 768P deliberately if you're iterating on a prompt and don't want to pay full price for throwaway tests.

Output is video, the generated clip, plus bizyair_model_name, an internal identifier tag for the model that served the request - useful for logging, not for wiring elsewhere in the graph.

Installing it

Search "BizyAir" in ComfyUI Manager and install, or clone it yourself: cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git, then restart. Set up your BizyAir API key on first launch - every node in the pack, including this one, is dead in the water without it.

Common issues

Because there's no negative prompt field here, if the model keeps generating something you don't want, your only lever is rewording the positive prompt to be more specific about what should and shouldn't happen - there's nothing to explicitly exclude with. Text-to-video also tends to wander more than image-to-video on composition and subject consistency, simply because there's no anchor frame; if you need tighter control over what the video starts from, BizyAir's Kling and Sora nodes in this same pack are image-to-video and give you that anchor. And as with every API node here, a failed or stuck generation is worth checking against your account balance before you assume it's a workflow problem - these are metered calls, not local jobs.