Happy Horse / v1.0 / textToVideo

This is the plainest node in the Happy Horse v1.0 family: a prompt, a few settings, and a video comes back - generated on Civitai's servers, not your machine. It's part of civitai/civitai-comfy-nodes, Civitai's official ComfyUI pack for its Orchestration API, so it needs no GPU, downloads no model, and bills the job in Buzz. If you've got a Civitai account and some Buzz to spend, this is a zero-friction way to generate short text-to-video clips without owning the hardware a video model would normally demand.

Why reach for it over a local video model? Because there's no "over" - this one exists entirely server-side. The local alternative (Wan, LTX, Hunyuan) means multi-GB downloads, VRAM gymnastics, and minutes per clip on a card that can handle it. Happy Horse is the opposite trade: you trade money for convenience and near-instant setup. It also slots into a bigger graph like any other node, so you can chain it after a prompt-enhancement node or before a local upscaler - the classic "closed model inside my workflow" move that the KB's external-API-nodes doc describes.

How it works

Mechanically it's a thin HTTP client with a nice face. The node assembles a videoGen workflow with the discriminator happyHorse / v1.0 / textToVideo, POSTs it to Civitai's orchestration endpoint, then long-polls until the job reports succeeded. When it does, the returned MP4 is downloaded into ComfyUI's temp directory and surfaced as a native VIDEO output. The prompt you type goes straight to Civitai's fleet - that's the whole mechanism, so treat the prompt box like a public chat window rather than a local GPU.

Inputs and outputs that matter

• prompt (required) - describe the whole clip. Camera moves, subject, lighting, mood. Video models punish vague prompts, and cloud ones are no exception.
• resolution - 720p or 1080p, default 1080p.
• duration - 3 to 15 seconds, default 5. Longer clips cost more.
• seed - set it to lock a result you like; 0 (default) means random each run.
• aspect_ratio - 16:9, 9:16, 1:1, 4:3, 3:4.
• api_config - leave it empty unless you're forcing a specific auth; it defaults to CIVITAI_API_TOKEN or your stored OAuth login.

Outputs are the pack standard: video (the clip), workflow_id (the orchestration job id), and raw_json (the full job response, handy when a run fails and you want the server's error verbatim).

Installing it

One install gets the whole Civitai pack. ComfyUI Manager → Custom Nodes Manager → "Civitai Comfy Nodes" (publisher civitai) → Install → restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

That's it - the only dependency is requests, and there are no model files to fetch.

Common issues

• "No credentials" on first run. Connect a Civitai Auth node or set CIVITAI_API_TOKEN in your environment. On a remote/headless ComfyUI, the env var is the way to go; interactive browser login won't work there.
• It costs Buzz every run. Video generation is the expensive end of the Civitai cloud. Each node shows the workflow id and cost in its status; check it after a few runs so the bill doesn't surprise you.
• Early preview caveat. The pack is explicitly flagged as still under active development - APIs and node shapes can change without notice. Don't build a production workflow on it without pinning the version.
• Blank prompt is caught locally. The node validates required text fields before submitting, so an empty prompt fails fast with a clear message instead of a server 400.