RunPod Kling v2.1 I2V Pro

Kling is the closed video service people point at in "what tool made this?" threads, and this node is Kling v2.1's I2V Pro tier as a RunPod public endpoint (kling-v2-1-i2v-pro). You give it an image, a prompt, a duration, and a movement budget - it gives you back a video URL. No Kling account, no separate web tool, no local video model eating your VRAM. If Kling's motion quality is what you want and you don't want to leave ComfyUI to get it, this is the node.

The "Pro" in the name mostly means the endpoint is Kling's stronger image-to-video tier, and the honest take is that I2V is where Kling shines anyway - anchoring the first frame to a known image solves most of the consistency problems that plague pure text-to-video. You bring the image, Kling brings the motion.

How it works

The pack's standard async plumbing, pointed at a video endpoint: POST your image (as a data URL) and prompt to RunPod, poll until the job completes, and get back a video URL string rather than a tensor. Your wired input_image becomes the payload image; the image text field is there for URL-based input instead of a wire. The video URL output is the whole deliverable.

The inputs that matter

• input_image - the frame to animate, from a Load Image node. The single most important input on this node.
• prompt - the motion/action description: "A majestic dragon breathing fire over an ancient castle" is the pack's default. Describe what happens, not just what the image shows.
• negative_prompt - filled in and useful; Kling has real CFG.
• guidance_scale - default 0.5, range 0–1. This is the creative-vs-adherence knob; the low default is intentional, Kling likes a light touch.
• duration - 1–10 seconds, default 5. Longer costs more and risks drift.
• enable_safety_checker - on by default, enforced server-side.

What comes out

A single video_url output (STRING). Download it right away - RunPod's video URLs expire after 7 days. Wire it to a Load Video from URL node or Save Text to keep the link.

Installing it

One of 27 nodes in tcarwash/ComfyUI_RunpodNodes. ComfyUI Manager: search ComfyUI_RunpodNodes, or

cd ComfyUI/custom_nodes
git clone https://github.com/tcarwash/ComfyUI_RunpodNodes

then restart. No extra dependencies - the pack only needs what ComfyUI already ships. Grab a RunPod key at runpod.io.

Gotchas

• Video is the expensive corner - per-video billing is an order of magnitude above per-image, and longer durations multiply it. Be deliberate.
• 7-day URL expiry - download or lose it.
• Queue times are real - video jobs queue and render server-side; 1–5 minutes is normal. The console's [RunPod] Status: polling log is your progress.
• Closed model, no bypass - moderation at the source, no weights to patch.

Kling v2.1 is a solid mid-2020s I2V tier - not the absolute frontier, but reliably good motion. This node makes it a three-wire addition to any workflow: image in, prompt in, video URL out.