HappyHorse 1.1 · Text to Video

Text-to-video is the hardest of the three video tasks and the one people reach for last, because a single prompt has to invent the subject, the wardrobe, the set, the light, and then animate all of it coherently. That's why the standard advice everywhere - including the local Wan ecosystem, where the image-to-video workflow is the community favourite for exactly this reason - is to anchor on a frame if you can. But sometimes you don't have a frame. You have a sentence, and you want the sentence to move.

HappyHorse 1.1 text-to-video takes a written brief and returns 3 to 15 seconds of clip at 720p or 1080p with generated sound, in any of nine aspect ratios from 21:9 down through square to 9:21. It is Alibaba's video model reached through SpicyAPI's API, so your GPU is not involved - you need an internet connection, a key, and a bit of budget.

Inputs

• prompt - multiline, and the only creative control on the node. Because nothing is anchored, be over-specific: subject, wardrobe, environment, light, camera, and the sequence of action. "A woman walks through a market" will get you a mush of averages. "A woman in a red raincoat walks left to right past fish stalls, camera tracking at chest height, steam and warm lantern light" gives it decisions to make.
• aspect_ratio - default 16:9, nine options including the tall social shapes. Set this first; it changes framing, not just cropping.
• resolution - 720p or 1080p, defaulting to 720p.
• duration_seconds - 3 to 15, integers. Longer clips cost proportionally more.

Note what's not here: there's no seed widget on this node. You can't lock an output and grind it. If you want reproducibility, use the image-to-video variant, where a seed is exposed and a fixed opening frame does the same job much better anyway.

Outputs

video and task id. Wire the first into Save Video; keep the second - it's the SpicyAPI job id, and it's the string support wants if something comes back broken.

Install and key setup

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

Restart ComfyUI. The pack needs no model downloads and no extra dependencies - it ships as pure Python over what ComfyUI already has. Then grab a key from spicyapi.ai/console/keys, paste it into Settings > SpicyAPI, and restart once so the node list is built from your account's catalogue. Server users: set SPICY_API_KEY in the environment. The key lives in your ComfyUI user folder, never in a node widget, so it doesn't travel with workflows or in image metadata.

If you don't see this node in the menu at all, restart - the catalogue is resolved at startup, so models you gained access to since the last boot only appear after a restart.

What actually bites

Per-second pricing and no seed. At about $0.14/s, a 15-second run is a couple of dollars, and without a seed you'll be paying each time you tweak a word. Prompt test at 3 seconds, in 720p, and only go long once the short clip does what you want. Set Max cost per run in the SpicyAPI settings so a stray 15-second experiment can't clear your balance.

Cancelling isn't cancelling. Interrupting in ComfyUI stops you waiting; the job keeps running on SpicyAPI and is charged if it succeeds. Its output stays in your console for up to 14 days.

Errors worth knowing. No SpicyAPI key is set - add the key. 40310 - verify your email. 40004 - no deployment serves that combination; change the parameter the message names. Any error ending in request_id or a task id is support material.

And a sanity note on what this is. Everything Alibaba shipped after Wan 2.2's open weights went API-only, so models in this tier are reachable this way or not at all. If you'd rather own the pipeline, a local Wan 2.2 image-to-video graph is free forever after the download - it's just older, and slower, and yours.