Kwaivgi Kling V3.0 Std First Last Frame (Channel Edition)

If you've got a beginning and an end already decided, you don't want a model inventing them - you want it to bridge them. That's what this node is for: feed Kling V3.0 Std your first frame and your last frame, and it generates the motion in between. It's the FLF (first-last-frame) mode, and it's the most underrated trick in the video toolbox for shots that need to land on a specific beat.

What it is

Another entry in BizyAirPlus's Kling channel - the "Std" tier of Kling V3.0, reached as a cloud model rather than local weights (Kling never shipped open). Where the T2V sibling starts from nothing but a prompt, this one starts from two images. The generation itself happens on BizyAir's cloud: no GPU load, no model files, just an API key from bizyair.ai and a request.

The workflow that shines here: generate your opening frame and your closing frame as images (Kling V3.0 supports image-to-image, or use any image model you like), then let FLF turn the space between into video. Character stays consistent because the model isn't guessing what the start looks like - it's interpolating with both endpoints pinned.

How it works

First-last-frame is a conditioning problem: the video model gets the start frame as a hard condition and the end frame as a target, then generates the sequence that plausibly moves from A to B. The catch, and it's a real one, is that plausibly is doing a lot of work - if your two frames disagree on lighting, wardrobe, or framing, Kling will happily mutate the subject to make the journey work. Keep the endpoints visually compatible and you'll get clean transitions; throw in a costume change and you'll watch the model do increasingly desperate things to reconcile it.

Inputs that matter

• images - your endpoint frames. The widget accepts a stack; the system adapts to the image ratio, so don't overthink resolution.
• duration - 3–15 seconds, default 5. Longer gives the motion more room but more chances to drift.
• prompt - describe the movement between the frames. "Camera pushes in, she turns to the window" beats leaving it blank.
• seed - lock it to reproduce a take; -1 for random.
• sound - on by default; Kling V3 generates audio with the clip.

api_config and skip_error are the same as every BizyAir node: per-request credential override, and fail-with-placeholder instead of aborting the workflow.

Outputs

video feeds your SaveVideo/preview chain. request_id is the job id for billing or support. urls gives direct download links if you want the file outside ComfyUI.

Installing

ComfyUI Manager → search "BizyAirPlus", or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart, then turn the BizyAirPlus button ON and enter your API key when prompted (stored at ~/.BizyAirPlus/apikey.ini, or via BIZYAIR_API_KEY).

Where people get burned

Endpoint drift is the #1 complaint with any FLF node - mismatched frames make the model contort. Fix it upstream: generate the two frames from the same seed/character reference. Also note this is the paid Std tier; each request costs credits, so lock a seed and validate your prompt on short durations before committing to the full 15-second version. If a job errors with no visible cause, the console usually has the real reason - network or key issues top the list.