PVL Kontext Max Single (fal.ai)

If you want the best single-image instruction editing the Flux family offers, and you want it without running anything locally, this is the node: PVL Kontext Max Single wraps fal's flux-pro/kontext/max endpoint. Give it an image and a sentence, get back the edited image. It's the one-image counterpart to the pack's Kontext Max Multi node - same flagship model, fewer inputs to juggle.

It's part of the ComfyUI Assistant Node pack from pvlprk, and mechanically it's the simplest of the Kontext wrappers: image in, prompt in, edit out. Kontext Max is API-only from Black Forest Labs, so the node isn't a shortcut around a local model - it is the way you use it from ComfyUI.

How it works

Your image is base64-encoded and submitted with prompt to the Kontext Max endpoint, then the result is polled for and pulled back as an IMAGE tensor. The prompt is an instruction - "change the sky to sunset," "turn the background to a city street" - and the model edits the input to match while preserving the subject. The in-context architecture is why it holds identity well across edits, which is the Kontext family's whole selling point.

The controls are the Pro-tier set: CFG (guidance scale, default 3.5), seed, num_images (1–4), output_format, sync_mode, safety_tolerance (1–6, default 2), and aspect_ratio - the same 21:9 → 9:21 menu, defaulting to 1:1. Note the default: unlike the Dev-tier Kontext node, match_input isn't an option here, so pick the shape you want the edited result to take.

The inputs that matter

• image - the input.
• prompt - the edit instruction.
• aspect_ratio - the output shape (default 1:1).
• CFG - 3.5 default; nudge for how strictly the model follows the instruction versus the input.

Output: one IMAGE.

Installing it

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

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

Set FAL_KEY (or FAL_API_KEY / FAL_CLIENT_KEY), restart, add credits.

Where people get burned

Because this is the Pro tier, cost and latency are higher than the Dev node - don't use it to iterate on prompts when the Dev node will get you 80% of the way for less. And watch for the pack's classic silent-failure pattern: errors can surface as a blank placeholder image, so glance at the console before you blame the model. If your edits come back too loose, raise CFG a touch; too tight and stuck to the input, lower it.