Kling v2.6 Pro (BudgetPixel)

Kling v2.6 Pro is the "begin here, end there" model. It does text-to-video, plus image-to-video with a start frame and an optional end frame - which is the trick that separates it from the newer Kling 3.0 nodes on this shelf, most of which only take a start frame. An end frame lets you pin down the destination: start with a character at a door, end with them across the room, and the model has to choreograph a believable path between the two instead of freewheeling. That's a real creative control, and it's the reason this older model still earns its 150 credits per second.

It also has generate_audio - optional audio, which is a contrast with the Turbo tier that bundles sound in unconditionally. If you want control over whether the clip comes out silent or with audio, v2.6 Pro gives you the switch.

How it works

Standard BudgetPixel video flow: your prompt, optional start frame, optional end frame, and the audio toggle go into the Kling pipeline, the node polls until succeeded, and the finished MP4 downloads back as a core VIDEO. The two-frame setup is the mechanical heart of the node: image seeds the first frame, end_image seeds the last, and everything between is the model's problem to solve. Skip end_image and you're doing ordinary start-frame image-to-video.

The inputs that matter

• prompt - required, multiline.
• image - optional start frame.
• end_image - optional end frame. The node's distinguishing feature; this is where the "compose a journey" workflow lives.
• generate_audio - the audio toggle. On for a clip with sound, off for a silent render to soundtrack later.
• length_seconds - your duration choice (per-second billing at 150 credits/sec - the Cost Estimate node is your friend).
• aspect_ratio - framing.

Output is a single VIDEO.

Install

Manager (search "BudgetPixel") or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

Restart, set BUDGETPIXEL_API_KEY or budgetpixel.json ({"api_key": "bpx_live_..."}) in the pack folder. No model files; requests is the only dependency.

Where people get burned

The classic mistake is treating end_image as a suggestion. It's a hard constraint, and if your end frame is wildly inconsistent with the start (different lighting, different clothes, different character), the model will do visible violence trying to reconcile them - the two frames need to look like two moments of the same scene, or the transition shows it. Second, this is v2.6, not v3: the newer Kling 3.0 tiers offer higher per-second ceilings and native audio by default, so v2.6 Pro is specifically the value pick when you need the end-frame control that v3 Standard/Turbo dropped. Third, with audio on, check the sound before you build a soundtrack around it - generated audio is decent, not bespoke.