mochi

If every other video node in the civitai-comfy-nodes pack feels like a cockpit, this one is a single button. Mochi is Genmo's video model, and this Civitai recipe node runs it entirely in the cloud: you type a prompt, hit run, and a clip comes back as a native VIDEO - no GPU work on your side, no model files, billed in Buzz like everything else in this pack.

That makes it the node to reach for when you want a quick text-to-video test and don't want to think. There's no aspect ratio dropdown, no resolution picker, no sampler. It's genuinely thin, and the thinness is the feature.

The inputs

The full input list fits on a Post-it:

• prompt (required, multiline) - this is all you must fill in. Since there are no resolution or style controls to lean on, the prompt is doing more work than usual here. Describe motion explicitly: camera movement, subject action, "slow zoom," scene changes. Mochi's an older generation of model, so it rewards plain, literal descriptions over prose.
• seed (required, default -1) - random by default. Set it to a positive value to reproduce a run; -1 rolls a new seed each time.
• enable_prompt_enhancer (optional, on by default) - rewrites your prompt server-side before generation. Leave it on for loose prompts; turn it off when you've written something specific and want it untouched.
• api_config - the auth socket, shared across the pack.

That's it. No image input here - this is text-to-video only.

Outputs

Standard pack trio: video (VIDEO) to wire into a Save Video / preview node, workflow_id (STRING) for tracking the job in Civitai's sidebar, and raw_json (STRING) for the full response if you want to see exactly what a run billed.

Install and auth

Same one-time setup as every node in this pack. ComfyUI Manager → search Civitai Comfy Nodes → install, restart. Or:

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

Then authenticate once: wire a Civitai Auth node, set CIVITAI_API_TOKEN, or sign in from the Civitai sidebar. The pack has no heavy dependencies - that requirements file really is one line.

When this is (and isn't) the right node

The honest take: Mochi is not the quality leader in this pack. If you want the current-generation stuff - MiniMax's Hailuo, Veo 3, Seedance - those are separate nodes with more controls and better output. This node's job is cheap, frictionless iteration: draft a concept, check the motion reads, re-roll. Because there are no controls, it's also the most beginner-proof node here - you can't misconfigure it, only write a bad prompt.

Watch the same gotchas as the rest of the pack: cloud generation is metered, the default timeout is 30 minutes (raise it via the Civitai Auth node or CIVITAI_COMFY_TIMEOUT), and Civitai's web-generator content filter applies here too. If you leave enable_prompt_enhancer on and get output that doesn't match what you typed, that's the enhancer rewriting you - flip it off and compare.

One more thing: the pack is an early preview and its nodes are generated straight from Civitai's API spec, so Mochi's capabilities may grow (or shrink) between releases. Check the node's fields after updates - a node this minimal can pick up new inputs without warning.