KIE Flux 2 (Image-to-Image)

Flux is the model line the open-source world grew up on - but the "2" generation, like the GPT Image and Grok models in this pack, is a cloud-only API product from Black Forest Labs. KIE Flux 2 (Image-to-Image) is the pack's way in: feed it one or more images and an edit prompt, and it returns a reworked image from either the Pro or Flex variant. If you've been running Flux locally since the 1.x days, this is the unromantic truth that the 2.x generation lives on someone else's GPU now.

The interesting bit is the batch side: images accepts 1–8 inputs and uploads them all. You can pass several references for a subject and an edit instruction, rather than one lonely source image. It's the same "small dataset in, one image out" idea as the GPT Image 2 I2I node, just with the Flux flavor and a couple fewer reference slots.

How it works

Standard KIE image job flow: upload the image batch, POST a createTask with the Flux 2 model id, prompt, and settings, poll recordInfo every 10 seconds (300-second timeout, auto-retry on transient errors), then download the result as a ComfyUI IMAGE. The model dropdown maps directly to KIE's model strings, so it stays current with what the marketplace serves.

The inputs that matter

• images - required. 1–8 source images; all upload. Batch order is preserved, so if you're referencing "the second image" in the prompt, know your ordering.
• prompt - required, 3–5000 chars. With multiple references, spell out what each one contributes.
• model - flux-2/pro-image-to-image (default) or flux-2/flex-image-to-image. Pro is the quality pick; Flex is the cheaper/faster draft mode. Worth knowing the difference before you spend - Pro is the one people actually complain about the price of.
• aspect_ratio - 1:1 default; options 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, plus auto.
• resolution - 1K (default) or 2K.

Output: image, feeding any IMAGE consumer - save, upscale, or the pack's I2V nodes.

Installing

Installs as part of the whole pack. ComfyUI Manager: search "ComfyUI Kie API". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/gateway/ComfyUI-Kie-API
# restart ComfyUI

No model files, no heavy dependencies. API key at config/kie_key.txt in the pack folder (copy from kie_key.example.txt).

Where people get burned

The Pro/Flex price gap is the first surprise - Pro image-to-image is one of the pricier seats in this pack, and drafting on Pro is how balances evaporate. Flex exists for exactly that reason; use it. Batch ordering bites again: with several reference images, there's no in-node preview of slot order, so build your batches deliberately. And the model list isn't frozen - KIE's catalog shifts over time, so if the dropdown's options ever disagree with reality, the fix is on the marketplace side, and kie.ai/logs will show you the server-side error when a job dies.