Juggernaut Base Flux

Juggernaut is the name that defined realistic SDXL - seventeen versions over twenty months, the default photorealism answer for most of the community's SDXL era, co-published with the cloud company RunDiffusion. This node is a different chapter: Juggernaut on Flux, hosted on Runware's GPUs, with no checkpoint download and no VRAM. If you remember Juggernaut as the model that made skin look like skin, this is the same brand applied to the newer architecture.

Why reach for it? You want Juggernaut's photography-first character - plausible skin, natural lighting, that "it's a photo" default - without committing 12GB of VRAM or an afternoon to download a checkpoint. It's also a full-featured node: the base Flux model with the complete Runware parameter surface, which makes it the Juggernaut node to start with.

How it works

Standard Runware diffusion node. positivePrompt, width, height (128–2048, step 64) go in; rundiffusion:120@100 runs as taskType: imageInference on Runware's GPUs; the image comes back as a native image (IMAGE) tensor.

The inputs a beginner actually touches:

• negativePrompt - what to exclude. Note the tooltip: it's ignored when guidance is off (CFGScale ≤ 1), which is a Flux-ism worth remembering.
• seed - 0 lets the API roll its own; set it for reproducible runs.
• steps / steps_value - the gate-toggle pattern: the box is off (model default) until you flip it and set 1–50 steps.
• CFGScale / CFGScale_value - same pattern, 0–20.
• numberResults - up to 20 seeded variations.

Then there's the whole stack of builder sockets - seedImage, maskImage, lora, controlNet, ipAdapters, pulid, ultralytics, outpaint, acceleratorOptions - each wired from the pack's Runware/Params builder nodes. That's where this becomes a real workflow node rather than a text-to-image box: chain a LoRA, feed reference images, inpaint with maskImage and maskMargin.

Output: image (IMAGE). The advanced_json slot takes hiresFix for a two-pass upscale.

Installing

Pack-wide: ComfyUI Manager → Custom Nodes Manager → search Runware → install → restart, or:

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

No model downloads - that's the point. API key from the Runware dashboard, via Settings, RUNWARE_API_KEY, or runware auth login.

Gotchas

Two traps. First, the gate-toggle inputs (steps, CFGScale, strength) are off by default on purpose - the model has usage rules, and the node leaves fields unset until you flip them so a default request never violates them. If your output looks wrong, check whether you actually enabled the settings you think you set. Second, cost: every queue is a paid render, numberResults multiplies it, and the title bar shows per-run cost. And if you want the fast lane, the Juggernaut Lightning Flux node is the distilled 4-step sibling - different checkpoint, same pack.