FLUX.1 [schnell] (custom model)

This is the "any FLUX checkpoint, hosted" node. RunwareArch_flux_1_schnell is a custom-model node whose model field defaults to FLUX.1 [schnell] (runware:100@1) but accepts the AIR of any checkpoint on the FLUX.1 architecture - dev, community fine-tunes, whatever the catalog holds. One node, the whole FLUX.1 family, no 12GB downloads per flavor.

The default is a smart one. FLUX.1 [schnell] is Black Forest Labs' distilled fast rung - "schnell" is literally German for "fast" - built to run in a handful of steps, which is why the whole architecture's community settled on it as the cheap, quick FLUX. Via Runware that speed means cheaper cloud runs too. If you've never touched FLUX because the hardware floor scared you off, this node is the lowest-friction on-ramp in the pack: pick a checkpoint from the search button, type a prompt, get an IMAGE.

How it works

imageInference with whatever AIR sits in the model field. Prompt and IMAGE inputs go up as base64, the sampling settings pass through, the image comes back as a native tensor. Because it's the FLUX architecture, the sampling vocabulary is FLUX-native: trueCFGScale (a separate true-CFG control FLUX uses, distinct from the CFGScale gate), promptWeighting (FLUX 1 uses sdEmbeds syntax rather than SD's compel), and - notable for a distilled model - steps and CFGScale are both gated off by default, letting the checkpoint's own defaults rule until you intervene.

The inputs that matter

• model (required) - the AIR, default runware:100@1. Use the search button to browse hosted FLUX.1 checkpoints.
• positivePrompt (required) - FLUX.1 is a strong prompt follower; say what you mean.
• width / height - 1024×1024 defaults, 128–2048, step 64 (the FLUX tile size - straying off-grid can produce seams).
• trueCFGScale - the FLUX-family guidance control; higher = stricter prompt adherence, lower = freer, more "painterly." Off by default.
• seedImage / maskImage - img2img and inpainting, with maskMargin.
• lora - the RUNWARE_LORA socket for the FLUX LoRA ecosystem.
• advanced_json - documented for hiresFix.

Installing it

Install the pack. 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

Key from runware.ai/api-keys → ComfyUI Settings → Runware API key (or RUNWARE_API_KEY).

Where people get burned

• Schnell is schnell. The default checkpoint is distilled for speed, and distilled models lose a little ceiling versus dev - that's the deal, and it's why the node lets you swap in another AIR.
• Step 64 resolution. FLUX works on 64px tiles; an off-grid width/height is how you get weird edge artifacts. Keep it aligned.
• Every checkpoint is a different price. The catalog's bigger fine-tunes cost more per run than the 4-step schnell default; the title-bar cost tells you what you're spending.
• Not every AIR resolves. The search button shows what's actually hosted; a random CivitAI ID that isn't served will 404.

FLUX.1 [schnell] custom model is the pack's FLUX gateway: cheap default, whole-family flexibility, and the FLUX-native knobs (trueCFGScale, tile-step sizing) explained by tooltips instead of forum archaeology.