FLUX.2 Max · Edit

Small surprise in the price list that's worth leading with: FLUX.2 Max · Edit costs about 11.42 credits ($0.1142) an image - noticeably less than the Flex edit node in the same pack at 17.31. If you assumed the "Max" in the name meant the most expensive click on the canvas, it doesn't. Max is BFL's top-quality hosted tier; Flex is their flexible tier, and flexible has never been free.

What you're buying is the best light and the best colour in the family. The pack's summary is "top-tier FLUX: photoreal light and exact brand colors", and on edits that combination is the whole story: the model's job is to make the new element sit inside the existing photograph, with the same lens behaviour, the same falloff, the same white balance. Klein gets close most of the time. Max is the one you use when "most of the time" isn't the standard.

Three references, one instruction

image_urls is Autogrow: connect an image, another socket appears, three is the ceiling. This is the field that decides what kind of edit you're doing:

• one source, and you're asking for a change to it;
• two sources, and you're compositing - person from one, room from the other;
• three, and you're doing product work where the shape, the material and the environment all come from different places.

The service addresses them positionally as image 1, image 2, image 3, so spell the mapping out in the prompt. One caveat that catches people wiring batches: every image inside a connected batch is treated as its own reference, so four frames in one socket plus a second socket is five references into a three-slot model - the API rejects it with a 400 naming the field.

Prompting a paid retouch

prompt takes 5,000 characters and the encoder is a Qwen3 LLM, so treat it like a note to a retoucher rather than a tag dump. Name the change, then name the invariants: Replace the label on the bottle with the artwork from image 2, keep the glass reflections, keep the shadow direction, keep the background. Explicit ownership, spatial relations, and no ((weighting)) - that syntax is disabled for this encoder and always has been for Flux 2.

aspect_ratio defaults to auto, holding the shape of your first source. Don't override it unless you're deliberately reframing. resolution is 1k or 2k, and the editing endpoints bill a flat rate instead of a size tier, so at the current rate 2k costs the same as 1k - take the resolution.

seed sits on fixed so a re-queue replays the cached result instead of charging again. Flip control after generate to randomize when you actually intend to buy another attempt.

Outputs

images - a normal IMAGE batch, ready for Save Image or a colour-match pass against your source. task id - string; the job's name on PopNow's side, for the API console and support.

Install

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

Manager → Custom Nodes Manager → PopNow → Install → restart does the same thing. Then a key from popnow.ai/api/console/keys into Settings > PopNow - or export POPNOW_API_KEY on a headless box - and restart once more so the node list rebuilds from your account. Nothing to pip install; the pack's dependency list is empty, but it does need a ComfyUI new enough for the V3 node API.

Where it's worth it, where it isn't

Worth it: brand work where a colour has to survive comparison with a style guide; product compositing where the light has to match or the shot reads as fake; a final pass on an image you'll actually publish. Not worth it: anything exploratory, and anything you'd describe as "I just want to see what happens." That's the 2-cent Klein node's job.

And no, there's no loras socket - Max takes no user weights here, same as Pro and Flex. Your adapters live on the Klein LoRA lanes.

Last thing: set Settings > PopNow > Max credits per run before you start. At eleven cents a click, the guard rail isn't paranoia, it's the difference between an afternoon of iteration and an afternoon of invoicing.