Flux Kontext Pro (Channel Edition)

This is the flagship: Kontext Pro, the instruction-based editor that made "change this image with a sentence" feel effortless, hosted by SiliconFlow as a single node. Feed it one image and an instruction, get back the edited image. The reason people care about Kontext over older approaches is architectural - your reference image is concatenated into the model's context alongside the prompt rather than injected as adapter embeddings, which is why identity and structure survive edits instead of drifting into something that merely resembles the source.

The display name says "Channel Edition," and that's the honest part: this is a provisioned channel of the hosted Pro tier, the one BFL never open-weighted. You can't download Kontext Pro. This node is your path to it, at the cost of cloud credits and the round-trip latency of sending an image up and pulling the edit back down.

Inputs and outputs

• images - required, a single IMAGE socket (unlike the Max I2I variant, which grows to accept multiple). Feed it the picture you want changed.
• prompt - required. The instruction. Precise sentences win: "replace the parked car with a vintage bicycle and keep the lighting identical" is the kind of instruction this model thrives on.
• aspect_ratio - dropdown, default 16:9. Note the difference from the siblings: the Pro I2I channel defaults to 16:9, not 21:9 - the edit channel presumes a conventional frame.

Outputs: image, request_id, urls (JSON string of output file URLs).

How it works

The standard BizyAir machinery - serialize image and prompt, submit to flux-kontext-pro-base/image-to-image, poll, download. The key comes from BizyAirPlus settings or BIZYAIR_API_KEY. Nothing runs locally; this is a pure cloud call, and its value is entirely in the quality of the hosted Pro tier plus the zero-setup story.

Where people get burned

• The known Kontext weaknesses apply. The KB's community findings carry over to the hosted tier: anatomy degrades at full-body scale, and big edits can skew proportions (head too big, torso too small). If you're editing full-body shots, expect to retry or work in tighter crops.
• One image, one edit. No autogrow here - if your workflow needs multiple references, the Max I2I node is the one to reach for; Pro I2I takes a single input.
• Failure looks like a red placeholder when skip_error is on. The console and the urls output carry the real error.

Verdict

If you're doing serious instruction-based editing and have the credits, Pro I2I is the premium you'd actually pay for - it's the tier that normally requires a BFL API account, now sitting in your ComfyUI graph. If your edits are casual or you're running a local Kontext Dev setup you're happy with, the upgrade is marginal for the price. The demo everyone runs first - watermark removal - works, and it's the fastest way to feel what Pro-level instruction editing does to a stubborn edit.

Install

With the pack:

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

Restart ComfyUI, confirm BizyAirPlus is ON and keyed, and give it a precise instruction. Precision is the whole game with this model.