HunyuanVideo 1.5 · Image to Video

The thing that jumps out here isn't the model, it's the price. From $0.024 per second of output - a five-second clip lands around twelve cents, eight seconds around nineteen - which makes this the cheap end of the video range in this pack. You can burn through twenty variations of a shot for the price of one clip on a premium tier. That is the entire reason to reach for it.

Why Hunyuan, of all the video models

HunyuanVideo is Tencent's open video family, and its local story has faded: the KB's long entry on it says the community moved to Wan 2.2 and LTX for general work, with Hunyuan 1.5 (the 8.3B, step-distilled version this node runs) arriving too late to win the ecosystem back. Local mentions collapsed to single digits a month by mid-2026.

That's exactly what makes a hosted node interesting: the model's two real strengths - it's fast, and it follows a prompt well for its compute class - without the weight download or adopting a model the tutorials stopped covering in 2025. Two caveats, though: the output is a silent 24 fps MP4 with no native audio pass, and clips are 5 or 8 seconds, full stop. Need sound baked in? Look at the Vidu or MiniMax nodes. Need twenty seconds? You're stitching.

The author also notes there's no content filter on this route, which is unusual enough to be worth knowing - Hunyuan's own reputation in the KB is a model that never shipped with sexual-content restrictions.

How it works

image_url is the opening frame, a real ComfyUI IMAGE socket that takes the output of Load Image - or any node upstream that emits an image. The clip starts from that frame and keeps its aspect ratio, so resolution is picking a quality tier (480p or 720p) rather than a shape. Your still is encoded and uploaded when the node runs; the pack caches uploads per account for about twelve hours, so a re-run with different settings doesn't re-send the same picture.

The prompt is optional, which is the nicest thing on this node. Working from a photograph you want gently moved - a portrait, a product shot, a landscape - you can leave prompt empty and let the model animate the frame on its own. Type one in when you want a specific motion or camera move.

Inputs, and the two that matter

• image_url - the opening frame. Required, because this is image-to-video.
• prompt - optional motion and camera description. Empty is a legitimate choice.
• resolution - 480p or 720p.
• duration_seconds - 5 or 8. Billed per second, so this is the dial that moves your bill.
• seed - same seed and settings gives the same result. It ships on fixed control-after-generate so an accidental re-queue reuses the cached run rather than paying again.

Outputs: video (a VIDEO, so wire Save Video) and task id, the string for console logs or support.

Worth noting: 480p at five seconds costs almost nothing, which makes this a natural draft engine in the classic cheap-draft → pick → upscale loop.

Install

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

Restart ComfyUI, then Settings > SpicyAPI and paste a key from spicyapi.ai/console/keys. On a remote or headless machine, SPICY_API_KEY in the environment is cleaner. Restart once more after adding the key so the node menu is rebuilt from your account.

No models, no wheels: dependencies = [] in the project file, and it uses only aiohttp, torch, Pillow and PyAV, all of which ComfyUI already has. That's a rarer property than it sounds - most Comfy node packs install into the same Python environment and fight over the same pins, and this one doesn't enter that fight at all. ComfyUI Manager can install it by searching the pack title if you don't want to clone.

Where people get burned

• No GPU is needed, but no internet means no node. Every run goes out to api.spicyapi.ai; a proxy or firewall in the way shows up as a "may be intercepting the request" error, and HTTPS_PROXY is honoured if you've set one.
• Every run is billed. Seeds are fixed on purpose so a double-queue doesn't double-charge, and Settings > SpicyAPI > Max cost per run will refuse any run quoted above your ceiling before a cent moves.
• Cancel ≠ stop. Cancelling ends your wait; the job keeps running on SpicyAPI, is charged if it succeeds, and its result lives in your console for up to 14 days.
• A model missing from the menu means "restart." The node list is built from your account at startup. Without a key, you're seeing the bundled snapshot from the release.
• Common errors: No SpicyAPI key is set, 40310 for an unverified email, 40004 when a chosen option combination has no deployment, and unknown failures that end in request_id ... - that id is the thing support needs.

One honest note on the money: at this rate the temptation is to fire off ten clips per idea. That's the right instinct, and also how a $5 balance disappears in an afternoon. Set the per-run cap, then let it rip.