PVL Kontext Pro (fal.ai)
Kontext Pro, the API-only editor, from inside ComfyUI
- image
- IMAGE
Flux Kontext came in two tiers: Pro and Max shipped as API-only in May 2025, and the Dev weights went open a month later. If you want the top tier - the one that never ran locally, the best of the Kontext line - an API is the only way to touch it. This node is that: it wraps fal.ai's fal-ai/flux-pro/kontext endpoint so you can drive the Pro model from a ComfyUI graph. It's the same "feed an image and a sentence" instruction editing Kontext made famous - watermark removal, clothing swaps, background replacement - just on the bigger, non-open sibling.
If you already have the open Dev weights and a GPU that fits 12B, you probably don't need this. But if you're on a weak machine, renting cloud time anyway, or just want the Pro-tier result without babysitting weights, this is the cleanest path. It's part of the pvlprk/comfyui-pvl-api-nodes pack, so it shares the install and the "every run costs fal credits" reality with its siblings.
How it works
The node base64-encodes your image and POSTs it with the prompt and settings to https://fal.run/fal-ai/flux-pro/kontext. Then it does something worth knowing about: it actively checks the response for trouble. If the API flags has_nsfw_concepts, it raises an error. If the returned image comes back all-black (a common "filtered/failed" signal), it raises an error instead of silently handing you a black frame. That's more defensive than most nodes in this pack, and it saves you from debugging mysterious black outputs.
Inputs that matter
promptandimageare required - that's the whole editing interface.aspect_ratiogives you the full preset ladder (21:9, 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16, 9:21; default 1:1) - a Kontext Pro nicety you don't always get on the Dev nodes.safety_tolerance(1-6, default 2) is passed through to the endpoint; higher tolerances allow more permissive content, but the node will still refuse if the API reports NSFW concepts.CFG(3.5),num_images(1-4),seed(-1 random),output_format(jpeg default / png),sync_mode.
One IMAGE output.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/pvlprk/comfyui-pvl-api-nodes
Restart and set FAL_KEY in the environment that launches ComfyUI (fal.ai/dashboard). Standard for the pack.
Where people get burned
- Pro is a paid, hosted model - treat every run as money. There's no local fallback and no free tier behavior; a single edit is a single API charge.
- It's fully cloud-bound. No Dev weights, no offline mode, no "run it on my 3090" escape hatch. If fal is down, so is your edit.
- The NSFW guardrail is real and enforced.
safety_toleranceup doesn't bypasshas_nsfw_concepts- the node hard-fails when the API flags content. If that's your workflow's need, the KB's pointer to Qwen-Image-Edit (Apache 2.0, uncensored) is the usual recommendation. - Don't expect a mask. This is the pure instruction editor; for masked, region-limited edits reach for
PVL Kontext Dev Inpaintin the same pack.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| image | IMAGE | — | |
| seed | INT | -1-1–4294967295 | — |
| CFG | FLOAT | 3.51–20 | — |
| num_images | INT | 11–4 | — |
| output_format | COMBO | jpeg | 2 options: jpeg, png |
| sync_mode | BOOLEAN | false | — |
| safety_tolerance | COMBO | 2 | 6 options: 1, 2, 3, 4, 5, 6 |
| aspect_ratio | COMBO | 1:1 | 9 options: 21:9, 16:9, 4:3, 3:2, 1:1, 2:3, +3 |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |