FLUX.2 Pro · Edit

FLUX.2 Pro is the version of Flux 2 you can't download - the undistilled base that BFL only ever sold through an API - and Pro Edit is its image-editing face. About 8.67 credits ($0.0867) per image, which is three and a half Klein 9B edits. You don't reach for it to experiment. You reach for it when the output is going to a client who will notice that the logo colour drifted two shades.

Why it's worth the premium

Two things the cheap tiers can't promise. Colour fidelity: the pack's own description for this family is "photoreal images with readable text and hex color control", and hex control is a real distinction when you're matching a brand system rather than approximating it. Instruction adherence on busy prompts: the distilled Klein models are confident and fast, and they will quietly re-imagine the parts of your instruction they find boring. Pro follows the brief more literally - which is exactly what you want when the brief has seven constraints.

And the honest third reason: it hands you up to three source images to merge, with the constraints of a bigger model behind the merge. Composite work - put this person in that room, wearing this jacket - is where the per-image price stops looking silly.

image_urls and the three-slot limit

image_urls is an Autogrow socket. You get one image input, and each connection grows another, up to three. The service refers to your sources positionally as image 1, image 2, image 3, so write instructions that use those numbers rather than "the first picture" - Place the model from image 1 into the interior of image 2, matching the light on her face to the window in image 3.

The trap worth flagging: a socket can carry a batch, and each image in that batch is sent as its own reference. Wire a four-image batch into one socket and you've submitted four references to a model that accepts three, and the run comes back as a 400 naming the field. Split batches upstream.

The rest of the form

prompt gets 5,000 characters and deserves all the care you'd give a written brief. Flux 2's encoder is a Qwen3 LLM reading instructions, so prose with clear ownership and spatial relations wins; prompt weighting is disabled for this encoder and masterpiece, best quality is an empty ritual. Say what changes and what must not change - "keep the face, keep the posture, keep the frame" is not wasted words.

aspect_ratio defaults to auto, holding the shape of your first source. Override it only when you're reframing on purpose. resolution is 1k or 2k, and the edit endpoints in this pack bill a flat rate rather than a size tier - so taking 2k currently costs nothing extra. The node's badge is the live answer if that changes.

seed starts fixed. Re-queue the same graph and you get the cached result instead of a second charge, which at this price is a feature you'll appreciate. control after generate → randomize is the switch for a genuinely new attempt.

Outputs

images is a standard IMAGE batch - send it to Save Image, or into a colour-match pass against your source before you show anyone. task id is a string; it's the identifier for the API console and for support.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui
# restart ComfyUI

Or Manager > Custom Nodes Manager → search PopNow → Install → restart. Then a key from popnow.ai/api/console/keys into Settings > PopNow (or POPNOW_API_KEY in the environment for a headless install), and restart again so the node list is rebuilt from your account. There's nothing to install alongside - the pack's dependency declaration is empty - but it does need a ComfyUI recent enough for the V3 node API.

Fine print

No loras socket on this node. Pro doesn't take user weights here; if you need your own adapter, you're on the Klein LoRA lanes, not this one.

Pro edit's output still isn't a guarantee. It keeps colour and follows instructions better, and it still refuses what the hosted model refuses - content_rejected is a policy wall, not a settings problem, and no amount of prompt rewording gets around a filter. Failed runs are refunded; read the request id and move on. And put a ceiling in Settings > PopNow > Max credits per run before you start queueing: nine cents a shot is fine, nine cents × forty exploratory shots is a different conversation.