Image models · Black Forest Labs

FLUX.2 [dev]

Name: FLUX.2 [dev]
Author: Black Forest Labs

Backbone: 32.2B
Native output: 1024×1024
Default steps: 28
License: flux-dev-non-commercial
Released: 2025-11

How to run it

Quantized files · GGUF for ComfyUI

Smaller GGUF versions for ComfyUI (with the ComfyUI-GGUF node). No sign-in needed; the model’s license still applies.

huggingface.co/unsloth/FLUX.2-dev-GGUF (opens in a new tab)

Original weights · Hugging Face

Full-precision files from Black Forest Labs, for fine-tuning, vLLM, Transformers, Diffusers or ComfyUI.

huggingface.co/black-forest-labs/FLUX.2-dev (opens in a new tab)

Free Hugging Face account needed: accept the license on the model page and the download starts right away.

Official repo · GitHub

Reference code, inference scripts and docs from Black Forest Labs. The weights themselves download from Hugging Face.

github.com/black-forest-labs/flux2 (opens in a new tab)

git clone https://github.com/black-forest-labs/flux2

Links checked 2 days ago against Hugging Face, Ollama and GitHub.

Apps that run FLUX.2 [dev]

Our pick for each kind of computer, assuming the model fits in its memory. Pick your own GPU or device in the Model Finder for an exact answer.

Mac (Apple silicon)

Draw Things

Native Mac app with its own fast Metal engine and quantized downloads.

Also good:

Windows PC, NVIDIA GPU

ComfyUI

Runs this model with built-in or popular custom nodes; GGUF files load with ComfyUI-GGUF.

Also good:

Linux, NVIDIA or AMD GPU

ComfyUI

Runs this model with built-in or popular custom nodes; GGUF files load with ComfyUI-GGUF.

Also good:

App	Type	Opens it as	Systems	Engine	Start
Unsloth Desktop	Desktop app	GGUF	macOSWindowsLinux	llama.cpp, MLX, stable-diffusion.cpp	Open in Unsloth Desktop
ComfyUI	Desktop app	Original or GGUF	macOSWindowsLinux	PyTorch, GGUF via ComfyUI-GGUF	Get ComfyUI
Draw Things	Desktop app	Built-in support	macOSiOS	Draw Things	Open in Draw Things
Invoke	Desktop app	Built-in support	macOSWindowsLinux	PyTorch	Get Invoke
SwarmUI	Desktop app	Original or GGUF	macOSWindowsLinux	ComfyUI	Get SwarmUI
Forge Neo	Desktop app	Built-in support	WindowsLinux	PyTorch	Get Forge Neo
SD.Next	Desktop app	Original or GGUF	macOSWindowsLinux	PyTorch	Get SD.Next
Stability Matrix	Desktop app	Original or GGUF	macOSWindowsLinux	installs ComfyUI, Forge, InvokeAI, SwarmUI, SD.Next or Fooocus	Get Stability Matrix
Krita AI Diffusion	Desktop app	Built-in support	macOSWindowsLinux	ComfyUI	Get Krita AI Diffusion

Which apps can open a model depends on the files published for it: GGUF for llama.cpp-based apps, MLX for Apple’s engine, the original weights for vLLM and SGLang. Compare every app.

Memory needed

Peak GPU memory at the native output size. Offloading moves the text encoder, or the whole pipeline, to system RAM between steps.

Precision	Everything on GPU	Text encoder offloaded	Sequential offload	Download
`FP16`	116 GB	67.9 GB	25.8 GB	Original weights (opens in a new tab)
`FP8`	59.7 GB	35.7 GB	14.6 GB	Original weights (opens in a new tab)
`Q8 GGUF`	61.9 GB	37.9 GB	15.4 GB	GGUF (opens in a new tab)
`Q4 GGUF`	46.8 GB	22.8 GB	10.1 GB	GGUF (opens in a new tab)

Components

Backbone: 32.2B
VAE: 0.084B
Mistral Small 3.1 24B: 24B

Best hardware for FLUX.2 [dev]

The least expensive GPU in our catalog that runs it without offloading is the Intel Arc B580 (launch price $249) at Q4 GGUF.

Quality	Hardware	Quant	Memory	Speed	Runs
90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA RTX PRO 6000 Blackwell	`FP8`	59.7 GB / 96.0 GB	24 s	Runs well
85Excellent quality (A+) · 85 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GeForce RTX 5090	`Q4 GGUF`	22.8 GB / 32.0 GB	38 s	Runs wellencoder in RAM
92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA B200	`FP16`	116 GB / 180 GB	22 s	Runs well
92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H200 141GB SXM	`FP16`	116 GB / 141 GB	25 s	Runs well
90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H100 80GB SXM	`FP8`	59.7 GB / 80.0 GB	23 s	Runs well
90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H800 80GB SXM	`FP8`	59.7 GB / 80.0 GB	23 s	Runs well
92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H200 NVL	`FP16`	116 GB / 141 GB	27 s	Runs well
90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GH200 Grace Hopper (96 GB GPU)	`FP8`	59.7 GB / 96.0 GB	23 s	Runs well
90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H100 NVL	`FP8`	59.7 GB / 94.0 GB	24 s	Runs well
90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H100 80GB PCIe	`FP8`	59.7 GB / 80.0 GB	28 s	Runs well

Check it on your own hardware

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.