Kling 2.5 Turbo Text-to-Video

No image, no problem. This is Kling 2.5 Turbo for when your starting point is just words: describe the scene and motion, and WaveSpeed's cloud renders the clip. It's the text-to-video companion to the 2.5 Turbo image-to-video node, and it carries one surprise in the spec sheet - only the Pro variant is offered on this endpoint. The standard tier exists for image-to-video but not here, so "Turbo = cheap" has a ceiling: you're getting the quality tier, just not the newest model.

It comes from the ERPK Collection, the community pack that wraps the WaveSpeed AI API alongside OpenAI, Claude, Gemini, and Grok. Cloud-billed, no local weights, nothing to download.

How it works

Type a prompt (max 2500 chars), pick aspect ratio, duration, and prompt adherence, and the node sends the job to api.wavespeed.ai, polling every 10 seconds until done. The output is a video_url STRING - preview and download it with the pack's Preview Anything node. The only hardware doing work on your side is the network card.

Note the seed widget's real role here: this endpoint has no seed parameter, so a fixed seed is cache control - it makes ComfyUI serve the cached clip and skip re-billing. Randomize it to force a fresh generation.

The inputs that matter

• prompt - everything. Scene and motion, max 2500 chars. There's no reference image to bail you out, so be concrete about subjects, camera, and action; vague prompts return vague clips.
• aspect_ratio - the usual combo, 16:9 by default. Match it to where the video will live.
• duration - 5 or 10 seconds. Start at 5; ten seconds is a lot of runway for a model to wander.
• guidance_scale - 0.0–1.0, default 0.5. Crank toward the top when the output keeps drifting from your prompt. The tooltip says it best: higher values reduce creative deviation.
• negative_prompt - elements to suppress. Handy for recurring artifacts, less useful than on image models but worth keeping in your back pocket.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt

Or search erpk in ComfyUI Manager and install ERPK Custom Nodes. No model downloads. You need a WaveSpeed API key with credits, set via right-click canvas > ERPK Settings (Settings > ERPK > API Keys) - stored per-user, never in the workflow.

Common issues

• Empty video_url after polling - the generation failed server-side. Check WaveSpeed's dashboard; content filtering is the usual suspect, and the node itself just relays the failure.
• "Why is my fixed seed clip never changing?" - because a fixed seed reuses the cached result on this endpoint. That's a feature (skip re-billing) if you want it, a trap if you don't. Randomize to re-run.
• Clips that ignore the prompt - raise guidance_scale. If it still ignores you, simplify the prompt; Turbo isn't the model for dense, multi-subject instructions.

The honest recommendation: if you have a reference image, use the image-to-video variant of 2.5 Turbo - first frames anchor the result better than any prompt. Text-only is the right tool when you genuinely have no visual to start from.