Kling Image to Video

Kling Image to Video is the pack's animator: feed it an image, and Kling's model makes the scene move. This is the node most people install this pack for - the "give me a still, get a clip" move that Kling does well, with the full model roster (v3, v2.6, v2-master, v1.6), duration options, optional end-frame, optional camera control, and AI-generated sound. It's the direct-API version of something you'd otherwise pay a web UI for, sitting right in your ComfyUI graph.

The positioning honestly matters here. Kling is a closed Kuaishou model - no open weights, no local run, metered per-call credits. This node is the "model you can't download" path from the KB's API-wrapper pattern: your image leaves the machine, Kling animates it, and the result lands back in the graph as a normal IMAGE frame batch plus an audio output. If you're already committed to Kling's ecosystem, this is the smoothest on-ramp in the pack.

The inputs that matter

• auth - the KLING_AUTH object.
• image - your starting frame. The quality ceiling of the whole clip lives here; a clean, well-composed still animates far better than a cluttered one.
• prompt - optional motion direction ("waves roll in, gulls cross the sky"). Kling will invent motion even with an empty prompt, but describing it makes the clip yours.
• negative_prompt - things to avoid.
• model_name - kling-v3 (newest) through kling-v1-6. v3 is the default and generally the right call; v2-master was the cinematic favorite before it.
• duration - 5, 10, or 15 seconds.
• mode - pro (higher quality, more credits) or std (faster, cheaper). Iterate on std, ship on pro.
• sound - BOOLEAN, default on. Kling generates ambient audio for the clip; flip off for silent or to keep costs down.
• cfg_scale - 0..1, default 0.5. Higher = tighter prompt adherence, lower = more creative drift.
• image_tail (optional) - an end frame. Give it one and Kling animates between the start and end images, which is exactly what the pack's dedicated Keyframe Video node wraps.
• camera_control (optional) - a KLING_CAMERA object from Camera Preset / Camera Control, for a directed move.

Outputs: video (IMAGE batch of frames), video_file (saved path), audio (AUDIO), url, task_id.

How to get a clip that doesn't disappoint

• Start from a good still. This can't be overstressed: I2V inherits the source. A generated image in a Kling-native aspect ratio (see the Aspect Ratio Picker) is the safest starting point.
• Describe motion, not the scene. The image already is the scene. Your prompt's job is "what happens," not "what's here."
• 5 seconds first. Longer durations cost more and expose more failure modes; nail the 5s loop, then extend.
• Sound on, then decide. Generated audio is a nice freebie on short clips; on longer renders, silence is cheaper and you can add real audio later.

Common issues

• Subtle or static motion - usually a too-vague prompt or a scene with nothing to move. Add explicit motion verbs, or pick a subject with inherent movement (water, hair, crowds).
• Credit sticker shock - 15s pro is a lot of credits. Run drafts at 5s/std and use the Cost Estimator to see the spread before the final.
• End-frame surprises - image_tail makes the model interpolate to it; if the end frame is wildly different from the start, the middle gets unstable. Keep them related.

Install

Ships in ComfyUI-Kling-Direct:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-Kling-Direct

or ComfyUI Manager → search "Kling Direct" → install → restart. No models, no GPU, no extra dependencies. Set up the Auth node once, and this becomes the most-used node in your Kling workflows.