FLUX.1 Krea [dev]

FLUX.1 Krea [dev] is the model where Krea's aesthetic taste got applied to Flux, and it's the node in this pack with the most complete toolset bolted on. Krea is the San Francisco lab that ran a closed generation service for years; in mid-2025 BFL shipped this Krea-tuned Flux checkpoint, and the result became a favorite for people who wanted Flux's reliability with a more polished, "aesthetic" default look. This node runs that exact model on Runware's cloud - and unlike the other FLUX.1 siblings in the pack, it exposes nearly every trick in the book.

The difference is in the sockets. Look at the optional inputs: seedImage, maskImage, outpaint, lora, controlNet, ipAdapters, pulid, ultralytics, acceleratorOptions - plus layerDiffuse for transparent output and strength/maskMargin for edit control. This is a full editing workstation, not a txt2img node. It's the one you reach for when your workflow needs image-to-image, inpainting, outpainting, or LoRA conditioning in a single cloud node.

What you set

Required: positivePrompt, width, height (128–2048, stepped by 64). Then the ones a beginner actually touches:

• seedImage - the base image for img2img. Wire one in and your prompt transforms it rather than dreaming from noise.
• maskImage - for inpainting: pair with maskMargin (32–128 extra context pixels around the mask, so the model has room to blend).
• strength - a toggle (flip it on) plus strength_value (0–1, default 0.8). How far the transformation goes from your input; lower = more influence from the original.
• lora / controlNet / ipAdapters - the builder sockets. These take typed values from the pack's Runware/Params builder nodes and are where LoRA and ControlNet conditioning enter your workflow.
• negativePrompt, steps (gated toggle), scheduler (76 options), CFGScale/trueCFGScale (gated) - the sampling dials, all defaulting to "(default)" or off, which is right.
• promptWeighting - (default) or sdEmbeds syntax for prompt-weighting conventions.
• layerDiffuse - flip on for transparent (RGBA) generation.
• advanced_json - documented for hiresFix.

Output is one image.

Why it stands out

Krea's model was the "make it look good without thinking" pick - the aesthetic tuning means its default output is more polished than base Flux. Combined with this node's full socket suite, it's the one cloud node that can stand in for a whole local editing graph. The honest caveat from the community's Krea coverage: the dev-tier weights ship under BFL's non-commercial license, so treat it as a hobbyist-quality pick and switch to a pro/commercial tier for client work.

Install

Standard pack install:

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

Restart, or install "Runware" from ComfyUI Manager; API key in Settings → Runware API key or RUNWARE_API_KEY. No weights download. Gotchas are the shared pack ones - cost per run on the title bar, gated toggles (flip strength on before its value does anything) - plus the licensing note above.