Wan 3.0 Pro · Text To Video

Here's a mild take: of the Wan 3.0 text-to-video tiers in this pack, this is the one most people should actually be running. Pro costs $0.144/s at 1080p against Pro Prime's $0.234/s - a 38% discount for a tier that is still very good, and 3.2× the base tier's $0.045/s. If you're producing clips you intend to keep rather than admire, the middle is where the value is.

What it does

Type a description of a shot, get 2 to 30 seconds of video back with audio in the same file, rendered on SpicyAPI's servers. No source image, no weights on your disk, no GPU. It's the same request body as Wan 3.0, just at the higher quality tier.

In the wider workflow this is usually the last mile: an image model makes a keyframe or a storyboard, a local or hosted LLM writes and refines the shot descriptions, and this node turns the winner into footage. Because it's a text node, it's also the cheapest thing to A/B against other hosted video families - same prompt, different node, compare the takes.

The inputs

• prompt - subject, motion, camera. Keep it about what happens, not about what things look like; the model fills the rest.
• resolution - 1080p by default, then 2k and 4k. Per the node's own tooltip, native Wan 3.0 renders cap at 1080p and the upper tiers are enlargements of that render. My advice: stay at 1080p and upscale yourself if you need the pixels, because you'll keep both the money and the control.
• aspect_ratio - adaptive by default, following the intent of the prompt; 16:9, 4:3, 1:1, 3:4 and 9:16 are there when you want to force a frame.
• duration_seconds - 5 by default, or -1 for smart duration so the model chooses a length from the prompt. That switch is worth playing with early, since a forced five seconds is what makes most prompts feel rushed.
• generate_audio - on; turn it off if you're laying music over the clip anyway.
• enable_prompt_expansion - on by default on the Pro tiers (the base 3.0 nodes leave it off). It expands your brief before generation, which helps thin prompts and obscures your own wording when you're debugging.
• seed.

Outputs are video - straight into Save Video - and task id, a STRING carrying the SpicyAPI task id for the console logs or a support email.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/spicyapi-ai/comfyui-spicyapi

ComfyUI Manager: search SpicyAPI. Restart ComfyUI, open Settings > SpicyAPI, paste a key from spicyapi.ai/console/keys, restart once more so the node list is rebuilt from your account. SPICY_API_KEY is the cleaner option on a server. Nothing else to install - no model files, and the pack ships an empty dependency list, leaning only on libraries ComfyUI already ships. Python 3.10+ and the V3 node API (tested against ComfyUI 0.39).

Money and mistakes

The node's badge shows the rate for the options you've picked, and Max cost per run in the SpicyAPI settings gets checked against a real quote before the request is submitted - anything above your ceiling stops there, uncharged. Do use it. Ten seconds at 1080p is $1.44, and it's easy to queue six variations while you're distracted.

Two behaviours worth knowing before they surprise you. Seeds default to fixed: queueing the same graph twice returns the first result rather than charging for a second, so flip control after generate to randomize when you want a new take. And cancelling only stops your side - the task finishes on SpicyAPI and is billed if it succeeds, with the result waiting in your console for up to two weeks.

Error messages are specific enough to act on: No SpicyAPI key is set, 40310 unverified email, 40004 no deployment serves that combination (change the option the message names), and an error tail with request_id ... that support can trace directly. A proxy or firewall warning means HTTPS_PROXY is set but not working.