Flux Kontext: Dev vs Pro, Licence, Price and ComfyUI Setup

FLUX.1 Kontext is a family of image models from Black Forest Labs that edits a picture from a written instruction: you supply an image, say what to change, and it returns the edited image. It comes as [dev], open weights you can run in ComfyUI under a non-commercial licence, and as [pro] and [max], hosted models priced at $0.04 and $0.08 an image.

Checked on 13 October 2026 against Black Forest Labs' Kontext announcement and [dev] release post, the model card, the flux repository, the Kontext docs, API pricing, release notes, the Non-Commercial License, the licensing page, its help article and the Usage Policy; and against ComfyUI's Kontext Dev tutorial, Partner Node tutorial and Flux examples. We have not timed or scored the models ourselves.

Quick answer

What is FLUX.1 Kontext?

Black Forest Labs announced FLUX.1 Kontext on 29 May 2025 as a suite of models that take both text and images as the prompt. A plain text-to-image model starts from nothing each time. Kontext starts from your picture and changes only what the instruction names. The announcement lists four capabilities:

• Character consistency: keeping a reference character or object recognisable across scenes.
• Local editing: changing one element without affecting the rest.
• Style reference: making a new scene in the style of a reference image.
• Iteration: building on the previous edit, one instruction at a time.

[pro] and [max] went live on the API that day. The open-weight [dev] model followed on 26 June 2025, described as a 12B parameter model that "can run on consumer hardware" and that "focuses exclusively on editing tasks".

Kontext is no longer the newest line. The Kontext docs open with a notice that FLUX.2 is now the recommended model for generation and editing and call Kontext [pro] "a previous-generation model", and the pricing page files it under previous generation. It is still sold. The FLUX family is set beside other image models in Midjourney vs FLUX and FLUX vs SDXL.

Kontext dev vs pro vs max

Prices are from the API pricing page, which bills in credits worth $0.01 each: 4 credits for [pro], 8 for [max], with the same price in the API and the Playground. The page publishes no free allowance. Through the API, the editing docs accept an input image of up to 20 MB or 20 megapixels, return outputs of about 1 megapixel in aspect ratios from 3:7 to 7:3, and deliver results through links valid for 10 minutes.

Is Flux Kontext free, and can you use it commercially?

The [dev] weights cost nothing to download, but the licence limits what you may do with them. The FLUX [dev] Non-Commercial License (v2.0, last updated 25 November 2025) names FLUX.1 Kontext [dev] and grants use "solely for your Non-Commercial Purposes". It defines those as personal research, testing, hobby projects, and evaluation by companies in a non-production environment, and says use "for revenue-generating activity" is not one of them.

The outputs clause is where people get confused. The licence says "You may use Output for any purpose (including for commercial purposes)", and the model card repeats it. But the same licence only lets you run the model for non-commercial purposes, and Black Forest Labs' help article lists Kontext [dev] as "commercial license required". The safe reading: an image you made while experimenting is yours to use, and making images as part of a business is not covered. The licence also requires content filters or manual review of outputs. This is a summary, not legal advice.

There are two ways to do paid work. The first is the API: the help article says "all usage includes full commercial rights with no separate license". The second is a weights licence. The licensing page lists Builder, Platform, Professional and Enterprise tiers built around the FLUX.2 models, with other models offered as add-ons. It shows no prices.

How to run Flux Kontext in ComfyUI

ComfyUI has supported Kontext [dev] natively since its release, and the official tutorial ships the workflow in the Template Library. No custom nodes are needed.

File names and folders are from ComfyUI's tutorial. Sizes are from Comfy Org's repackaged model, the flux_text_encoders repository and, for the VAE, Black Forest Labs' FLUX.1 [schnell] repository. You need one t5xxl file, not both. ComfyUI's examples page also names the original flux1-kontext-dev.safetensors from Black Forest Labs' repository, which is gated: you accept the licence on Hugging Face before the files appear.

Kontext [pro] and [max] also work inside ComfyUI, through Partner Nodes that call the hosted models and bill Comfy credits to your account. That tutorial adds an Image Stitch workflow that joins several pictures into one input. If you add community node packs for FLUX later, install them through ComfyUI Manager.

What hardware does Kontext [dev] need?

Black Forest Labs publishes no VRAM figure for [dev], only the phrase "consumer hardware". Plan from file sizes: 11.9 GB for the fp8 model, plus 5.16 GB or 9.79 GB of text encoder. ComfyUI's startup flags reference says weight offloading is on by default on Nvidia and AMD cards, so a model larger than your graphics memory can still run, more slowly. We give no generation times because no primary source publishes them.

How to write a Flux Kontext prompt

ComfyUI's tutorial lists the habits that work: be specific, edit in steps, and state what must stay unchanged. It adds a word choice: use "change" or "replace", not "transform".

• Simple change: "Change the car color to red."
• Outfit: "Change the jacket to a black denim jacket while keeping the same facial features, hairstyle and expression."
• Background: "Change the background to a beach while keeping the person in the exact same position, scale, and pose."
• Text on a sign: put the words in quotes, as in "Replace 'joy' with 'BFL'", and add "maintain the same font style".

Describe the person instead of using a pronoun. The tutorial's example is "The woman with short black hair" in place of "she". For a recurring persona, our guide to Flux Kontext for character editing covers the reference-image side, and the free AI influencer prompt generator drafts the first still.

Black Forest Labs lists the limits itself. Long multi-turn sessions "can introduce visual artifacts that degrade image quality", and its example shows visible damage after six edits in a row. The model sometimes ignores part of an instruction. When an image starts to degrade, go back to an earlier clean version.

Whose face are you editing?

An instruction editor will change the clothes, setting or expression of anyone in a photo. A persona is a fictional adult you created, or you. Editing or reproducing a real person's face without their written permission is not allowed, and the publisher's terms say so too. The licence forbids use that violates "rights of publicity or 'digital replica' rights", and the Usage Policy (revised 4 August 2026) bars unlawful use of "a real person's name, image, voice and/or likeness" and realistic depictions of a real person in an intimate, degrading or misleading manner without documented consent.

The model card says API outputs carry signed C2PA metadata showing they were made with the model, and the Usage Policy forbids removing it. Platforms have their own disclosure rules for realistic AI images; Instagram's are in our AI label rules guide.

Kontext vs FLUX.2, FLUX 3 and Qwen-Image-Edit

Inside Black Forest Labs' own range, Kontext is now the older option. FLUX.2 [pro] and [flex] launched on 25 November 2025 with editing from up to 8 reference images through the API and output up to 4 megapixels. FLUX.2 [klein] followed on 15 January 2026, with its 4B model under Apache 2.0. FLUX 3 Image arrived on 1 October 2026 and combines up to ten references. Kontext [pro] is not the cheapest way to edit on the API:

Outside the FLUX family, the closest open model is Qwen-Image-Edit from Alibaba's Qwen team. Its model cards carry the Apache 2.0 licence and list Chinese and English, where Kontext [dev] is non-commercial and English only. Its fp8 file in ComfyUI is 20.4 GB against 11.9 GB for Kontext. Those are published facts, not a quality verdict: the files and versions are in our Qwen-Image-Edit guide.

Flux LoRA training and Kontext

A LoRA is a small add-on file that teaches a base model a subject or style. The announcement calls [dev] "suitable for customization", and Hugging Face listed 246 adapter models built on it when we checked. The licence treats a fine-tune as a Derivative, under the same non-commercial terms as the base model.

For a persona, the usual order is to train the character LoRA on a text-to-image FLUX model, then use Kontext to vary the stills. Our Flux LoRA walkthrough covers that training run, for a fictional persona or yourself only. The AI Influencers program covers the whole pipeline, and the free guides are in the AI influencers library.

Flux Kontext: FAQ