FLUX, the image models from Black Forest Labs: the family, and where it stands
FLUX is a family of image generation and editing models developed by Black Forest Labs, and a major one for local generation. What the family covers, where it stands (FLUX.2 klein for retouching by area, FLUX 3 announced) and how Axiris follows it.
FLUX is the family of image models from Black Forest Labs, for generation and editing; it spans several generations and variants whose capabilities, licences and hardware requirements differ, and it holds a major place in local generation and ComfyUI workflows. At Axiris, FLUX.2 klein is the editing and retouch-by-area engine used in production, and FLUX 3, announced on 25 July 2026 with an open-weights Dev version planned, is followed in the news. Every test separates what Black Forest Labs announces from what we observe.
A major family in Axiris workflows
FLUX is a family of models developed by Black Forest Labs for image generation and editing. Since FLUX.1, the family has grown through several generations, variants and tools, aimed at different levels of quality, control and deployment.
Axiris follows this ecosystem particularly closely because of its place in local generation and in advanced workflows.
A family, not a single model
The name FLUX covers several generations and several variants. Capabilities, licences, hardware requirements and ways of working change considerably from one model to the next: you read the model card and the licence of the model you use, not those of the family (AI model licences).
This page is the entry point to the family. The versions that warrant it are documented separately.
Generation and editing
Depending on the version, FLUX plays a part in text-to-image, image editing, image to image, visual references, ComfyUI workflows, LoRAs, local pipelines and APIs. The publisher's documentation covers text-guided editing and multi-reference workflows.
Where the family stands
FLUX.2 klein is the model we use in production for editing and retouching by area: you select what has to change, the inference is applied there, the rest does not move. Black Forest Labs positions it as a compact model for generating and editing quickly, with variants for iteration, fine-tuning and local execution. Its role in our pipeline: Models of the moment.
FLUX 3 was announced on 25 July 2026: a multimodal model trained on image, video, audio and language; video first, in early access, image next, and a FLUX 3 Dev open-weights version officially planned. On the day of the announcement, nothing was known about its size, its memory requirements or its integration into ComfyUI. Our reading: FLUX 3 announced in Early Access.
FLUX in ComfyUI
FLUX models fit into ComfyUI graphs and combine with other operations, which allows pipelines far wider than plain generation:
references → FLUX → editing → mask → correction → upscaling
Where the files go and how to organise them: Models in ComfyUI.
FLUX at Axiris
We test FLUX generations in our image creation and transformation workflows. When we publish a test, we distinguish the capabilities announced by Black Forest Labs from the results observed in our own workflows, under known and dated conditions: Comparing AI image models. For editing engines as a whole: Image editing models.
Official sources
Black Forest Labs · Model catalogue · FLUX documentation
Contact request: a first conversation, free and with no commitment. Every practice is different; it is by listening to you that we propose what suits you.