flux2 / dev / createImage

Compare this node to its sibling for one second and the naming tells you everything. flux2 / createImage exposes model (pro/max) and nothing else. flux2 / dev / createImage has no model enum at all - it's the dev checkpoint, full stop - but hands you guidance_scale, num_inference_steps, and a loras socket. That's the difference between a closed API product and an open weights pipeline, and it's exactly the right way to think about this node: Flux 2 dev is the checkpoint you could download and run, and this is Civitai running it for you, billed in Buzz, with the sampler knobs you'd expect from a local workflow.

Why use it instead of the local install? The dev weights are the ones that need 18–24GB of VRAM even quantized - the hardware wall that kept Flux 2 a paper champion for most of the community. If you're on a machine without that, this node is the honest door: same model, your choice of steps and guidance, LoRA stacking through the Civitai LoRA Selector, and the result lands back in your graph as a native IMAGE tensor. The defaults are the sensible dev recipe: guidance_scale 2.5, 28 steps, 1024×1024.

The inputs that matter

• prompt - required.
• guidance_scale (2.5) - the dev-model guidance knob; low, as the family prefers. Below ~2 gets mushy, above ~4 gets harsh.
• num_inference_steps (28) - the sampler budget. 28 is the dev sweet spot; fewer is faster and softer, more is slower and barely better.
• loras - CIVITAI_LORAS socket. This is the big one that the pro/max node doesn't have: wire a Civitai LoRA Selector and stack Flux 2 dev LoRAs from Civitai's catalog without downloading anything.
• width / height - 1024×1024 default, up to 2048.
• enable_prompt_expansion - off by default; dev understands your prompt fine on its own.
• quantity, seed, output_format, image_metadata - standard.

Outputs: images (IMAGE), errors, workflow_id, raw_json.

Install and auth

The one pack install covers it. ComfyUI Manager → Civitai Comfy Nodes, or:

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

Authenticate (Civitai account with Buzz): export CIVITAI_API_TOKEN=..., a Civitai Auth node, or the sidebar sign-in.

Where people get burned

The most common mistake is expecting pro-tier output from dev-tier billing - dev is the open, tweakable checkpoint, and it's honestly a step below pro/max on the very hardest prompts (fine text, complex scenes), while being far more controllable. The second is under-guessing the cost of a full-res, 28-step dev render at quantity 4; the node shows per-job cost inline, so glance at it. And if you have the VRAM, the "should I just run this locally" question is real: dev is Apache-licensed and local gives you privacy and zero per-call cost. This node is the right call specifically when your hardware can't, or when you want Civitai's LoRA catalog wired in with zero file management - which, honestly, is a nicer workflow than most local LoRA installs.