Segmind Video Request

Same idea as this pack's Novita video node, different provider: SegmindVideoRequestNode sends an image and prompt to Segmind's hosted API and gets a video back, defaulting to wan-2.2-i2v-fast - the speed-optimized variant of Wan 2.2's image-to-video model. If Novita's pricing or queue doesn't work for you, this is the alternative cloud path to the same underlying model family.

What it's for

It's a request node, not a local model - you're calling Segmind's servers, not running anything on your own GPU. That makes it a reasonable pick when you want video generation folded into a larger ComfyUI graph without standing up a local Wan install, or when you specifically want the "fast" tier's speed/cost trade-off over a full-quality run.

How it works

You give it a hosted image URL, a text description of the video you want, and (optionally) generation parameters that map directly onto Wan's own settings - frame count, resolution, aspect ratio, and a sampling-schedule parameter - then it round-trips to Segmind and hands back the finished clip.

The inputs and outputs that matter

• image - the author's own tooltip on this field translates to "input image URL": like the Novita node in this pack, this wants a URL your source image is already hosted at, not a wired-in IMAGE tensor.
• prompt - described in the node's tooltip as the video description; this is your text prompt for the motion and content.
• model (default "wan-2.2-i2v-fast") - free-text field for Segmind's model name, so you're not locked to the default if Segmind adds other options.
• num_frames (default 81, minimum 81, max 100) - that 81-frame floor isn't arbitrary. It's Wan's native clip length; the model wasn't trained with a shorter context, so this node doesn't let you ask for less than what the underlying model actually supports.
• resolution (480p / 720p, default 480p) and aspect_ratio (16:9 / 9:16, default 16:9) - your output shape and quality tier.
• go_fast (default true) - an additional speed toggle layered on top of the already-fast model variant.
• sample_shift (default 12, range 1–20) - a flow-matching sampling-schedule parameter carried over from Wan's own generation settings; if you're not sure what it does, the default is a reasonable place to leave it rather than a value worth guessing at.
• frames_per_second (default 16, range 5–24), seed, api_key, timeout - round out the optional inputs.

Outputs: video (VIDEO), width, height, fps, and url (the hosted result location).

How to install it

Via ComfyUI Manager: search "ComfyUI-utils-nodes," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/zhangp365/ComfyUI-utils-nodes

Restart ComfyUI. No local model to download - you'll need a Segmind account and API key separately, which goes in the api_key field.

Common issues & troubleshooting

Same gotcha as the pack's other video-request nodes: image wants a URL string, not an IMAGE tensor. If you're feeding this straight off a LoadImage or a KSampler output, you need an image-hosting step in between - upload the result somewhere reachable and pass that URL in, rather than wiring pixels directly.

The num_frames floor of 81 is worth internalizing rather than fighting - trying to force a shorter clip isn't something this parameter supports, because it maps to a real limit in the model itself, not an arbitrary node restriction. If you want a shorter final video, trim it after generation rather than expecting this node to produce fewer frames. And as with any billed API node: an unset or invalid api_key fails every request, and timeout (default 300s) is the first thing to raise if higher-resolution or longer requests are dying before they complete rather than erroring cleanly.