PixVerse V6 Text-to-Video

What it is

PixVerse V6, text-only, as a ComfyUI node. Prompt in, clip out, with a duration dial that goes from one second to fifteen - which is a much more useful control than the fixed 5-or-10 second menu most hosted video endpoints give you.

PixVerse is one of the closed video services that turned up in the official ComfyUI Partner Node roster alongside Kling, Veo and Seedance (docs/knowledge/external-api-nodes.md). There are no weights, so the API is the only door. Why this node over Comfy's own: it's a separate path to a hosted model, billed through your KIE key rather than Comfy credits, and priced on KIE's marketplace. If you're using a few KIE models, keeping all of them on one key and one balance is worth something on its own.

Duration flexibility is where it separates from the pack's MiniMax and reference-heavy nodes: for a B-roll loop, a three-second beat between two shots, or a quick motion test, being able to ask for exactly three seconds is the difference between cheap and wasteful.

How it works

The node is generated from KIE's request schema for the PixVerse V6 text-to-video endpoint. Your prompt and settings are submitted as a task, the node polls until the job completes or timeout_seconds (1200 by default) expires, and the finished clip comes back as a ComfyUI VIDEO object. The task_id output is not throwaway - the PixVerse V6 Video Extension node in this pack wants exactly that value.

Inputs and outputs

Required:

• prompt (STRING) - the description of the shot, 3 to 5000 characters. The shipped default is a mist-shrouded mountain lake at sunrise; it's fine as a smoke test.
• aspect_ratio - 16:9 default, plus 4:3, 1:1, 3:4, 9:16, 2:3, 3:2, 21:9. The tooltip notes this applies to text-to-video and to the Fusion multi-reference mode.
• quality - 720p (default), 360p, 540p, 1080p. Your cost dial. 360p is genuinely useful for composition tests.
• duration - an INT from 1 to 15 seconds, default 5.

Optional:

• generate_audio_switch - off by default. Turn it on and the clip comes back with audio matched to the content.
• generate_multi_clip_switch - off by default; this is the multi-shot option. Leave it alone until you've seen a single-clip result you like.
• seed - 123456789 by default, range 0–2147483647. Same parameters plus same seed gets you closer to repeatable output.
• timeout_seconds, callback_url.

Outputs: video, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left.

Install

Manager: search KIE-Nodes-Next (author felipederosilva - there are several KIE packs in that search, so match the repo), install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next

Then Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. It's validated before it replaces a saved key, and you get a masked fingerprint plus your credit balance - the fastest way to tell "out of credits" apart from "something's broken". KIE_API_KEY in the environment takes priority for server installs.

Requirements are requests, Pillow, numpy, PyYAML. On Windows there's a one-click .vbs installer in dist/; pick a v0.3.1-or-later build, since the first release could leave backup folders inside custom_nodes that ComfyUI tried to import as nodes.

Where people get burned

Duration × quality is the bill. Fifteen seconds at 1080p is not the same purchase as three seconds at 360p. Test cheap, then spend.

generate_multi_clip_switch changes what you're buying. It's not "more of the same clip" - it's a different mode. Confirm it's what you wanted before you build a workflow around it.

Audio is a switch, not a quality level. There's no way to steer it beyond on/off here; if you need a specific track, generate silent and score it in post.

A 20-minute timeout on long jobs. Default timeout_seconds is 1200. A 15-second 1080p clip on a busy queue can brush up against that.

Endpoint churn. These are third-party marketplace endpoints; models get added and retired. If a node that worked stops resolving, run KIE • Refresh Model Catalog and let the pack re-read KIE's docs.

Compare against local before defaulting to this. Wan and LTX run on your own card for free after setup - hosted video wins on convenience and on specific capabilities, not on cost per clip.