Zhenzhen_kling_image2video

Kling (可灵) is the video model everyone compares everything to, and this node is the Comfly pack's image-to-video client for it. You feed a still, add a prompt, and a video comes back after a few minutes of relay time - the README is honest that Kling generations take roughly 5–6 minutes, so this is not a sit-and-watch node. The payoff: the model lineup here includes the recent Kling 2.x versions, and the camera-control options are the most elaborate of any Comfly video node.

Inputs that matter

• image - required. Your starting frame.
• image_tail - optional; an end frame to animate toward. Kling's best-known trick.
• prompt - required; describe the motion, not a still.
• negative_prompt - optional but worth using; Kling respects it.
• model_name - kling-v2-1, kling-v2-1-master, kling-v2-master, kling-v1-6, kling-v1-5, kling-v1. The master models cost more - the README's warning is literally "very expensive, use as needed."
• mode - std or pro. Read this one: the source shows v1.5 and v1.6 only support pro mode - pick std with those models and it won't generate.
• duration - 5 or 10 seconds.
• num_videos - 1–4 videos per run. The bill scales with the count, obviously.
• imagination - 0–1, default 0.5. How much creative license the model takes with your image; lower = stays closer to the source.
• camera - the fun one: none, horizontal, vertical, zoom, vertical_shake, horizontal_shake, rotate, and the master-only moves (master_down_zoom, master_zoom_up, master_right_rotate_zoom, master_left_rotate_zoom).
• camera_value - -10 to 10, the strength of that camera move.

Outputs: video (into ComfyUI's native Save Video), plus video_url, task_id, video_id, and response.

Install

Standard Comfly:

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

Or ComfyUI Manager → "Comfyui_Comfly" → restart. No local models; credits + key at https://ai.comfly.chat (Comflyapi.json or the node's apikey).

Gotchas

The std/pro trap is the #1 beginner stumble - the dropdown defaults to std, but for several models std silently won't produce output, so if you pick v1.5/v1.6 and get nothing, flip to pro. The master camera moves only exist on master models; pick a non-master model and those options are dead weight. And with 5–6 minute render times and paid slots, don't iterate by trial-and-error - lock the image, prompt, and camera first, then spend. The pack also has Comfly_kling_multi_image2video (up to 4 reference images, 1.6 model) if one still isn't enough.