FLUX.1 [schnell]

FLUX.1 [schnell] is the younger, faster sibling of FLUX.1 [dev], and in the Runware pack it's the budget pick for good reason. "Schnell" is German for "fast," and it's the distilled tier of the 12B family: designed to run in a handful of steps instead of the 20-50 the base model wanted. In the cloud, where you pay per generation, that's not just speed - it's the difference between a fraction of a cent per image and, well, a slightly bigger fraction of a cent. Runware's own numbers put FLUX Schnell at thousands of images per $10. It's genuinely the cheapest serious-quality image generation this pack offers.

There's a second reason people pick schnell over dev, and it's not about speed at all: licensing. Schnell is Apache 2.0, the permissive tier of the FLUX family, while dev is non-commercial. If you're doing anything commercial - freelance, product shots, monetized content - schnell is the FLUX you're allowed to use. The trade is the usual distillation deal: fewer steps means less refinement, so detail and prompt adherence trail dev slightly. For iteration-heavy work, that's the right trade.

What you set

The essentials are positivePrompt, width, and height (128–2048, step 64). After that, the settings are almost identical to dev, and the same guidance advice applies:

• CFGScale - off by default and should stay off. Schnell is distilled; the guidance is baked in.
• trueCFGScale - the real CFG knob if the model ignores you.
• steps - the whole point here is few steps. The default already handles this; adding steps buys little on a distilled model.
• seed, scheduler, numberResults (up to 20), negativePrompt (ignored at CFGScale ≤ 1) - all familiar.

Missing from schnell versus dev: no layerDiffuse (no transparent PNGs here) and no controlNet/pulid sockets - you get lora, ipAdapters, ultralytics, and the standard seedImage/maskImage inputs. That's fine, because you're not reaching for schnell when you need surgical control; you're reaching for it when you need 50 variations fast and cheap.

Output is image (IMAGE), straight into Preview Image or Save Image.

How it works

Same pack machinery as every Runware node: build a request with the model AIR and taskType, send it over REST, download the finished IMAGE back. The node's title bar shows cost per run, and with schnell it's the rare model where the number genuinely looks like a rounding error. You'll never load the 12B model locally, never manage the T5 encoder, never wonder if your VRAM fits it - the cloud does all of it.

Installing

ComfyUI Manager → search Runware → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Dependencies are light (runware-sdk, pillow, soundfile), no model downloads. You need a Runware API key via Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

Troubleshooting

Same pack gotchas: no key means the "No Runware API key" error, so set that first. If images look muddier than dev, that's the distillation tax, not a misconfiguration - you can nudge trueCFGScale up a touch if you must, but don't expect dev's crispness at schnell's price.