ComfyUI Node

Comflowy Hailuo

Hailuo image-to-video inside ComfyUI — closed model, zero VRAM

By 6174·Created 2 years ago·Updated 2 years ago· 17
Comflowy Hailuo
  • image
  • video
prompt
prompt_optimizertrue
seed0

Hailuo is MiniMax's image-to-video model - closed weights, genuinely good motion quality, and absolutely not something you can run on the 8GB card in your desktop. This node is Comflowy's way of putting it inside a ComfyUI graph anyway: you drop in an image, write a prompt, and the whole thing happens on Comflowy's servers. If you've been watching the "image-to-video" workflow trend and wishing you could use the hosted models without leaving ComfyUI, this is that.

The catch is in the README's price table: Hailuo runs about 5000 credits per video. That makes it the most expensive node in this pack by a mile - the image models cost hundreds, this costs thousands. It's a "I really want this specific motion" tool, not a "generate a dozen drafts" tool.

How it works

Standard Comflowy API node pattern: your image gets base64-encoded and POSTed with your prompt to Comflowy's cloud endpoint (this one routes through a different host than the default - Hailuo hits comflowy--comflowyspacecloud-web-main.modal.run), the service runs the video generation, and the result comes back as a downloaded .mp4 file in your ComfyUI output folder.

Here's the part that trips up first-timers: the output is a VIDEO - which in this pack means a file path string, not something you can preview directly. The README is explicit: this node requires the Comflowy Preview Video node wired to its output. Forget that and you'll see a path in your console and wonder where the video went.

The inputs

  • image (required) - the start frame, any standard image.
  • prompt (required) - what should happen. Motion description matters more than in image gen; think "camera slowly pushes in while the subject turns" rather than a static scene description.
  • prompt_optimizer - boolean, default true. Rewrites and enriches your prompt before sending it. Leave it on unless you're chasing a specific phrasing.
  • seed - 0 to 2^31-1, default 0. Standard reproducibility knob.

Output: video (VIDEO type).

Installing and the usual suspects

Install the pack via ComfyUI Manager (search "Comflowy") or git clone https://github.com/6174/comflowy-nodes into custom_nodes, restart, and set your key with Comflowy Set API Key. Dependency-wise it's just requests.

Troubleshooting, grounded in how the pack actually behaves:

  • No video appears. Check you've got Comflowy Preview Video on the video output. The node itself isn't an output node; something has to display or consume the file.
  • error.mp4 in your output folder. When an API call fails, the pack's base class returns a placeholder error.mp4 path instead of crashing your whole graph. If you see that file, the real error is in your ComfyUI console - usually a network problem or an expired/debited key. The README calls out network issues as the top cause of API failures.
  • Credits vanish. At ~5000 credits a pop, a couple of failed-looking runs that actually succeeded will show up as a drained balance. Set seed and check your console before re-running.

If you're already generating start frames locally with Flux or SDXL, the natural workflow is: generate image → this node → Preview Video. The API does the heavy lifting; you keep the graph.

CategoryComflowy

Inputs (4)

NameTypeDefaultDescription
imageIMAGE
promptSTRING
prompt_optimizerBOOLEANtrue
seedINT00–2147483647

Outputs (1)

NameTypeDescription
videoVIDEO