FLUX.2 Flex · Text to Image

Flex sits in the middle of Black Forest Labs' hosted tiers: pricier than Dev, cheaper than Max, and pitched at exactly one job - text and lettering that comes out legible instead of melting into squiggles. This node wraps that endpoint, and it's what I'd reach for when the image has to contain a word.

What Flex actually is

Flex is API-only. There are no Flex weights to download, no Civitai page, no GGUF quant waiting on Hugging Face - it exists as an endpoint on BFL's serving stack and PopNow resells it per image through your credits. So the honest framing is: you're not choosing between Flex and your local Flux. You're choosing between Flex and another hosted lane.

The pack's summary for it - "Spells headlines and labels exactly as written" - is the marketing line and also the actual use case. Multi-panel comic lettering, a product label, a street sign that needs to say something specific. Flux 2 got a lot better at glyphs in general; Flex is the tier where BFL is willing to charge for the consistency.

One thing you should know before you go looking: in BFL's original framing, Flex was the tier that exposed steps and guidance scale so you could trade speed against quality per request. This node does not expose them. Its schema is prompt, aspect_ratio, resolution, seed - the same four fields as every other text-to-image node in the pack. Whatever internal step schedule Flex runs, it's decided server-side. If fine-grained sampling control was what you wanted from the word "Flex", you're not getting it here.

The fields

prompt takes up to 5,000 characters. Flux 2's encoder is a Qwen3 LLM, so put the exact string you want rendered in quotes inside a normal sentence - a diner window with "OPEN 24 HOURS" painted in red - rather than stacking text, typography, logo tags. Since you're paying nine cents a try, write the whole brief in one go: subject, framing, light, and the literal copy.

aspect_ratio covers 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9 and 9:21. resolution is 1k or 2k at 2,048px on the long side, and on Flex the 2k step is a real jump in price - roughly 8.7 credits at 1k versus about 13 at 2k. For a headline test at 1k you'll already read the text fine.

seed starts fixed. That's a billing feature, not a creative one: queue the same graph twice and ComfyUI reuses the first result instead of buying a second. Flip control after generate to randomize when you're hunting for a take, and keep an eye on how many takes you're hunting.

Outputs

images is an ordinary IMAGE - Save Image, Preview Image, an upscaler, whatever your graph does next. task id is the string PopNow uses for the job; keep it around if you might need to look the run up in the API console later.

Install

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

Or through Manager > Custom Nodes Manager, search PopNow, Install, restart. Then a key from popnow.ai/api/console/keys into Settings > PopNow - and restart a second time, because the node list is rebuilt from your account's catalog at startup. The pack declares no dependencies at all; the only hard requirement is a ComfyUI new enough for the V3 node API that all these nodes are built on.

Fine print worth reading first

Hosted tiers move. The figures here come from the pack's own catalog and node descriptions at the time of writing; the badge on the node is the current rate and the quote before each run is the real number, so trust those over this page.

Flex is not the cheap iteration tier. At this rate, a ten-image exploration costs more than a Klein 9B session by a factor of five. My advice: nail the composition and wording on a Klein node, then spend one or two Flex generations on the final with the copy you've already proofed. And if the endpoint rejects your settings with PopNow 40004, it's naming a combination it won't serve - usually a ratio/size pairing - not something wrong with your prompt.