Local AI Local AI Intermediate

Run Flux Locally

Download and run Flux.1 [dev/schnell] on consumer hardware

64 of 66

What is Flux?

Flux.1 is an open-source image generation model from Black Forest Labs. It produces high-quality images with excellent prompt following and text rendering.

Official site: blackforestlabs.ai

Variants

VariantLicenseQualitySpeed
Flux.1 [dev]Non-commercialHighestSlower
Flux.1 [schnell]Apache 2.0Very highFastest

Minimum requirements

  • FP16: 16+ GB VRAM
  • NF4/GGUF quantized: 8–12 GB VRAM
  • CPU offload: works on less VRAM, much slower

Download Flux for ComfyUI

You need three files:

  1. UNET
ComfyUI/models/unet/flux1-dev.safetensors
  1. CLIP
ComfyUI/models/clip/flux_clip.safetensors
  1. VAE
ComfyUI/models/vae/flux_vae.safetensors

Search Hugging Face for ComfyUI-compatible Flux checkpoints.

Quantized versions

If you lack VRAM, download NF4 or GGUF versions of the UNET. Load them in ComfyUI with the corresponding loader nodes.

Prompt

Flux understands natural language well:

a futuristic Tokyo street at night, neon signs, wet pavement reflecting lights, cinematic

Tip

Schnell needs only 4 sampling steps. Dev needs 20–28. Adjust the KSampler accordingly.

Working out which model to run this on? See The Codex. Packaging it as a reusable skill? See The Armory.