Protoface Kling 3 Turbo Pro · Text to Video

Turbo is a workflow decision, not a spec

Every hosted video model line now has a fast tier and a quality tier, and the fast tier is the one that changes how you work. You can't iterate on a model that costs real money per look - so you write one prompt, queue it, and pray. A speed-optimised tier turns that into an actual loop: six takes, pick one, then re-render the keeper properly.

Kling 3.0 Turbo does 1080p text-to-video, optimised for speed. This node is that task in Protoface, the beta pack that runs hosted models as native ComfyUI nodes. Paid per generation, remote, no weights. Think of it as the draft tier of the Kling 3.0 line - same family as Kling 3.0 Pro, tuned the way distilled models in this ecosystem are always tuned: less sampling, faster answers, slightly less fidelity.

How it works

The pack generates this node from the provider's public schema and runs everything through one shared adapter: validate, upload, journal, submit, poll, download, native VIDEO. No per-model code, which is why the Kling Turbo nodes and the Kling Pro nodes are the same shape on the canvas with a shorter control list.

generation_id is the paid-request identity - fixed to reuse or resume, incremented to spend.

The inputs that matter

prompt is a media-style control on this node (Default / Set / Null), and the tooltip names the exclusivity: it's mutually exclusive with multi_prompt, and you provide exactly one of them. Set it for a single shot; reach for the multi-shot helper nodes when you want cuts - search the model name plus "Shot", connect one helper per shot, and set the list control to Set. Shot durations have to add up to duration.

duration defaults to 5 seconds, as a combo. aspect_ratio defaults to 16:9 - set vertical before you render, because reframing generated footage is a losing game.

cfg_scale (default 0.5, range 0–1) is Kling's adherence dial rather than a local sampler's CFG: higher values reduce flexibility and bind the output closer to your prompt, lower values give the model room. Start at the default and move one direction at a time.

multi_prompt is the shot list, and it's the reason this node has fewer controls than the Pro version: no negative prompt, no sound toggle. If you need those, you're on the wrong node.

Outputs

video (native VIDEO, list), files (PROTOFACE_FILE list of the originals), run id (string). Wire video into a preview or save node, or into another Protoface video input - the media inputs offer a References mode that accepts a reused result asset, so a turbo draft can feed a Pro re-render without touching disk.

Install

Manager → search Protoface → install protoface-nodes → restart → reload. Manual, from the ComfyUI directory:

git clone --branch v0.1.0 --depth 1 https://github.com/protoface-ai/protoface-comfyui.git custom_nodes/protoface
python -m pip install -r custom_nodes/protoface/requirements.txt

ComfyUI 0.41.0+ and Python 3.10+, installed with the interpreter your ComfyUI runs on. Then click Protoface in the lower-right and enter a key from your Protoface account - it's billed there, not against Comfy credits, which trips up people used to Partner Nodes.

Where people get burned

Shot durations that don't sum to duration. The multi-shot list is validated, not best-effort. Build the arithmetic first.

Expecting Pro controls on the Turbo node. No negative_prompt and no sound here. If you need to exclude something from the prompt, that's a Pro-tier control.

Queueing drafts without incrementing generation_id. Every take needs a bump. Requeueing identical inputs with a fixed id deliberately hands you the recorded job back.

Judging the model by a Turbo pass. Faster tier, same rough motion quality, less detail. If the composition works and the texture doesn't, that's exactly what the quality tier is for.