by Black Forest Labs
Next-generation text-to-image model from Black Forest Labs. 32B parameter DiT with Mistral-Small-3.2-24B text encoder. Requires ~64GB+ VRAM at full precision; consumer GPUs need Q4/Q8 quantization.
VRAM requirements, GPU fit, and setup notes for Flux.2 Dev, including 8GB/12GB fit guidance where relevant. Recommended runtimes: ComfyUI and Diffusers support. Best download size: ~11.9 GB at FP8.
Your hardware
Detecting...
Measured quality metrics for Flux.2 Dev outputs.
How often humans prefer this model's output (0-100%)
Visual quality and composition rating (5-9 scale)
Compare which GPUs can run Flux.2 Dev at different precisions. FP8 uses less memory than FP16 when available, and the grade shows how comfortably each GPU handles the workload.
| Resolution | VRAM Required | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 | 63.5 GB | F● | F● | F● | F● |
| 768×768 | 63.7 GB | F● | F● | F● | F● |
| 1024×1024 | 64.0 GB | F● | F● | F● | F● |
| Resolution | VRAM Required | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 | 19.7 GB | A● | F● | F● | D● |
| 768×768 | 19.8 GB | A● | F● | F● | D● |
| 1024×1024 | 20.0 GB | A● | F● | F● | F● |
Turbo / LCM distillation
Use distilled scheduler at 4-8 steps for faster iteration
ControlNets available
Add guided generation with 3 adapters (+3.6 GB VRAM each)
from diffusers import FluxPipeline
import torch
pipe = FluxPipeline.from_pretrained(
"black-forest-labs/FLUX.2-dev",
torch_dtype=torch.float16
)
pipe.to("cuda")
image = pipe(
prompt="your prompt here",
num_inference_steps=28,
guidance_scale=3.5,
height=1024,
width=1024,
).images[0]
image.save("output.png")Get started
Setup instructions for running Flux.2 Dev locally
1. Download the model
Get the checkpoint from HuggingFace
2. Place in:
ComfyUI/models/checkpoints/3. Launch ComfyUI
python main.pyVRAM allocation at 1024×1024 on RTX 4090 24GB (24 GB)
Time per image at 1024×1024, 28 steps, FP16.
Download Flux.2 Dev in the precision that matches your GPU. Lower precision usually means less VRAM pressure, while higher precision keeps more quality.
3 ControlNets available for Flux.2 Dev. ControlNets add guided image generation (edges, depth, pose) at the cost of extra VRAM.
Edge detection for structural guidance. FLUX.1 ControlNets are largely compatible with FLUX.2.
Depth-based spatial control. FLUX.1 ControlNets are largely compatible with FLUX.2.
Single model handling canny, depth, pose, tile, and blur. FLUX.1 ControlNets are largely compatible with FLUX.2.
Growing Flux 2 LoRA ecosystem
Frequently asked questions
Flux.2 Dev (32B parameters) requires approximately 64.0 GB of VRAM at FP16 precision for standard 1024×1024 image generation. If you want a lighter setup, lower precisions like FP8 can reduce memory use when available.
Flux.2 Dev usually needs more than 8GB for comfortable local use. Check the VRAM table above for the exact resolution and precision trade-off.
Flux.2 Dev is marked for ComfyUI and Diffusers support in our catalog, so those are the runtimes we recommend first for local setup. If your workflow uses another front end, check the model's available formats and workflow notes above before downloading.
Flux.2 Dev is too large for the RTX 4090's 24 GB VRAM at FP16. Consider using FP8 precision or a GPU with more VRAM.
Yes, Flux.2 Dev has 3 ControlNet adapters available: Canny Edge, Depth Map, Union (Multi-Control). Each ControlNet adds roughly 3.6 GB of extra VRAM.
Growing Flux 2 LoRA ecosystem The LoRA ecosystem for Flux.2 Dev is rated as "moderate". Each LoRA adds roughly 0.3 GB of extra VRAM.
On a reference GPU (RTX 4090 24GB), Flux.2 Dev generates a 1024×1024 image in approximately ~3m 9s at FP16 with 28 inference steps. Faster GPUs with higher memory bandwidth will produce images more quickly.
See also