flux2 / createImage

Flux 2 is, by broad agreement, the most capable open-weights image model there is - a 32B-class rectified flow transformer that does generation and editing in one checkpoint, with an unrivaled text encoder for in-image text. It's also a hardware monster that needs 18–24GB of VRAM even quantized, which is why most people never actually run it and why this node matters. flux2 / createImage hands you the hosted BFL model through Civitai's orchestration API - pick pro or max (the enum is right on the node, default pro), type a prompt, get an IMAGE tensor back, billed in Buzz. It's the strongest model in this pack, and it is the definitive case for the "model you can't run" half of the API-node story.

The node is deliberately plain, which is fine, because the model carries the weight. The inputs that matter:

• model - required: pro / max, default pro. Max is the top tier, pro is the sensible default; for most work pro is where the value is.
• prompt - required. Flux 2's text encoder is genuinely good at natural-language prompts and at rendering words inside the image - lean into descriptive sentences.
• width / height - 1024×1024 default, range 512–2048. Flux 2 handles big generations and is where its 4MP-class output shows.
• enable_prompt_expansion - off by default here (the reverse of the Qwen nodes). Leave it off; Flux 2's own understanding is the whole point.
• quantity, seed, output_format, image_metadata - standard.

Notably absent: no guidance_scale, no steps, no LoRA socket, no reference images. This is the hosted product API's createImage, and BFL keeps the internal knobs. If you want the sampler controls, the flux2 / dev sibling node in the same pack exposes guidance, steps, and LoRAs - that one runs the dev checkpoint and is a different animal.

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

Install and auth

The one-time pack install. 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.

The honest tradeoff

Flux 2 is the one node in this pack where the cloud call is easiest to justify: you simply cannot run pro/max class inference on consumer hardware, and renting 80GB-class GPUs per-image is more fiddly than per-call Buzz. The caveats are the standard cloud-API ones - your prompts leave the machine, the model's moderation is BFL's, and per-call cost on a top-tier model is not a "batch 50 thumbnails" budget item. If your job is high-stakes hero images or text-in-image work, this node is the strongest door in the pack. If you just need a fast draft, everything else is cheaper, and flux2-dev gives you the sampler knobs.