PVL FLUX DEV (fal.ai)

FLUX.1 Dev was the model that made the community say "this is what we wanted SD3 to be" - natural-language prompting that finally worked, hands and text that stopped melting. It's still a great everyday generator. But it's a 12B stack, and if you're on a card that can't hold it comfortably, you've got two options: fight with fp8/GGUF quantization, or let someone else run it. This node is the second option - a straightforward wrapper around fal.ai's flux/dev endpoint with no image inputs, no LoRA plumbing, no endpoint auto-detection. Just prompt, size, sampler knobs, generate.

It's the simplest node in the ComfyUI Assistant Node pack's fal family, which is both its charm and its limitation: for the basic "turn a prompt into an image in the cloud" job, there's nothing to get in your way.

How it works

Text-to-image only. Your prompt and settings go into a JSON payload, the node POSTs it to fal-ai/flux/dev with your key, and downloads the returned images as a ComfyUI IMAGE tensor. The parameters map straight to the API: steps → num_inference_steps (default 28), CFG → guidance_scale (default 3.5), plus seed, num_images (1–4), output_format, and sync_mode. The acceleration menu (none / regular / high) trades a little quality for speed - none is the default and the safest for consistent output.

One default worth noticing: width is 1392 and height is 752 - the classic FLUX.1 Dev landscape pair, the model's sweet spot. The ranges go 256–1440, so you can't accidentally request a 4096-wide canvas here; Flux.1 Dev's native training resolution is modest, and the node keeps you inside it.

The inputs that matter

• prompt - the whole game. FLUX.1 Dev reads natural language, not tag soup.
• width / height - defaults 1392×752; switch to portrait by flipping the values.
• steps / CFG - 28 and 3.5 are the community's standard starting points. Flux.1 Dev is forgiving; moving these is for when you're chasing a specific feel.
• seed - set a value for reproducible output, or leave at -1 for random.
• enable_safety_checker - on by default here (unlike some of the pack's other nodes), so be ready to turn it off if you're hitting the filter.

Output: one IMAGE.

Installing it

Part of pvlprk/comfyui-pvl-api-nodes. Install via ComfyUI Manager (search "ComfyUI Assistant Node") or:

cd ComfyUI/custom_nodes
git clone https://github.com/pvlprk/comfyui-pvl-api-nodes

Set your fal.ai key as FAL_KEY (the code also accepts FAL_API_KEY / FAL_CLIENT_KEY), restart ComfyUI, and top up credits.

Where people get burned

The pack-wide traps all apply - missing FAL_KEY, and errors that surface as a blank placeholder image with the real message in the console. One Dev-specific note: this node is the plain text-to-image wrapper, so if you wired up an image expecting an edit or a LoRA expecting style transfer, you're on the wrong node - that's what the pack's Flux Dev LoRA, Flux General, or Flux2 nodes are for. Use this one when you want exactly one thing, done simply.