☁️BizyAir Kling O1 Reference Video/Images To Video

What it is

This is the most reference-heavy node in the Kling section, and it's built for keeping specific subjects or a specific style consistent across a generated clip. You can feed it up to seven reference images plus a reference video simultaneously, all steering the same generation - think "keep this character's face and this outfit consistent" or "match the visual style of this clip" while the prompt describes what actually happens. That's a meaningfully different job from the plain text-to-video or single-image-to-video Kling nodes elsewhere in this pack, which have no equivalent multi-reference conditioning at all.

The inputs that matter

• prompt - what happens in the generated clip.
• duration - 3 to 10 seconds, and unlike the other Kling video nodes in this pack, this is a continuous integer range rather than a fixed 5-or-10 choice.
• aspect_ratio - 16:9, 9:16, or 1:1.
• keep_original_sound - off by default. If you're feeding in a reference video and want its original audio carried through to the output, turn this on.
• ref_image_1 through ref_image_7 (all optional) - up to seven reference images, whatever combination is relevant to your generation. None are required; use as many or as few as your reference needs actually call for.
• ref_video (optional) - a reference video, on top of or instead of the still-image references.

Outputs: video, and bizyair_model_name, reporting kling-o1 as the backend model resolved for this endpoint.

How to install it

ComfyUI Manager: search "BizyAir," install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAir.git

A BizyAir API key is required - this is entirely a hosted, closed-model call with no local fallback.

Common issues

Too many conflicting references muddy the result. Seven reference image slots is generous, but each one is influence competing with the others plus the prompt. If a result looks like an unclear compromise between references rather than a clean read on any of them, try trimming down to just the references that actually matter for that specific generation instead of filling every slot by default.

Original audio not carrying through from a reference video. Check keep_original_sound - it's off by default, so unless you deliberately turn it on, expect a fresh (or silent) audio track rather than the reference video's original sound.

Not sure whether to use this node or the plainer O1 image-to-video node. If you just need to animate from a single starting frame, the simpler I2V node is the right tool and will likely be cheaper per call. Reach for this one specifically when you need multiple references or a reference video's style/content carried into the generation - that's genuinely unique to this node in the pack.

Job fails or produces nothing. Check API key and account credits first, same as every other node in this section - with this many inputs and a heavier request, it's also worth confirming each reference image/video actually loaded correctly upstream before assuming the API call itself is the problem.