Nodes/ComfyUI Assistant Node/PVL Flux.2 Pro (fal.ai)
ComfyUI Node

PVL Flux.2 Pro (fal.ai)

The flagship, with none of the local install

By pvlprk·Created about a year ago·Updated 8 months ago· 1
PVL Flux.2 Pro (fal.ai)
  • image_1
  • image_2
  • image_3
  • image_4
  • image_5
  • image_6
  • image_7
  • image_8
  • IMAGE
prompt
image_sizecustom
width1024
height1024
safety_tolerance2
enable_safety_checkerfalse
seed-1
num_images1
output_formatpng
retries2
timeout_sec120
debug_logfalse
delimiter[++]

Flux 2 Pro is the top of the Flux 2 line - the one you can't download at all, because Black Forest Labs only serves it through an API. Which is exactly what this node is for. PVL Flux.2 Pro (fal.ai) wraps fal's flux-2-pro endpoint and drops it into your ComfyUI graph, so you get the flagship's quality without a local model, without VRAM pressure, and without having to leave ComfyUI to go use fal's playground. The trade, as always with these nodes, is a per-generation bill and a network round-trip.

It's part of the ComfyUI Assistant Node pack from pvlprk. If you've already installed that pack for the other fal nodes, this one is sitting there waiting - same key, same credits, same pattern.

How it works

The node auto-selects its endpoint from your wires: bare prompt → fal-ai/flux-2-pro, any of the up-to-8 image inputs connected → fal-ai/flux-2-pro/edit. Images are base64-encoded inline and the call goes to fal's queue API, which is polled until the result is ready and then pulled back as a standard IMAGE tensor. Prompts can be split on the delimiter (default [++]) and the pieces fanned out into parallel requests, each with its own retry budget.

Pro's character shows in what it hides: no steps, no CFG, no sampler knobs. The Pro tier is a black box - you set the size, the seed, the safety tolerance, and the output format, and the model decides how to get there. That's by design; it's the "just take the best result" option in the family.

The inputs that matter

  • prompt - the only creative input that matters.
  • image_size - presets (1:1, 3:4, 9:16, 16:9) or custom for width/height (512–2048), plus auto for the edit endpoint. Note there's no 4:3 preset here, unlike the Dev node - use custom.
  • safety_tolerance - 1–5, default 2. This is your main knob for how strict fal's filter is.
  • num_images - 1–8.
  • retries / timeout_sec / debug_log - the pack's standard reliability trio; timeout_sec defaults to 120 here, reflecting that Pro jobs run longer.

Output: one IMAGE.

Installing it

It's 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 (or FAL_API_KEY / FAL_CLIENT_KEY), restart ComfyUI, and make sure you have credits.

Where people get burned

Same pack-wide traps: unset FAL_KEY, and errors that show up as a blank placeholder image rather than a clean failure - check the console. Pro also costs more per call than Dev or Flex, so resist the urge to use it for iteration; iterate on Flex or Dev and spend Pro on the final render. And because Pro edits don't take a steps control, don't go hunting for a field that was never there.

CategoryPVL_tools_FAL

Inputs (21)

NameTypeDefaultDescription
promptSTRING
image_sizeCOMBOcustom6 options: 1:1, 3:4, 9:16, 16:9, auto, custom
widthINT1024512–2048
heightINT1024512–2048
safety_toleranceINT21–5
enable_safety_checkerBOOLEANfalse
seedINT-1-1–4294967295
num_imagesINT11–8
output_formatCOMBOpng2 options: jpg, png
retriesINT20–10
timeout_secINT1205–600
debug_logBOOLEANfalse
delimiteroptSTRING[++]
image_1optIMAGE
image_2optIMAGE
image_3optIMAGE
image_4optIMAGE
image_5optIMAGE
image_6optIMAGE
image_7optIMAGE
image_8optIMAGE

Outputs (1)

NameTypeDescription
IMAGEIMAGE