Flux.1 Dev in ComfyUI
12B rectified-flow transformer delivering top prompt adherence and typography when loaded via DualCLIPLoader (clip_l + t5xxl) and ae.safetensors. Requires 12 GB+ (8 GB with GGUF Q4 / FP8) VRAM and loads from models/unet/ or models/checkpoints/.
Updated Oct 17, 2026 · 1 min read
- Model Family
- Black Forest Labs
- VRAM Target
- 12 GB+ (8 GB with GGUF Q4 / FP8)
- ComfyUI Folder
- models/unet/ or models/checkpoints/
- Precision
- FP16, FP8 e4m3fn, or GGUF Q4_K_S / Q8_0
- Default Sampler
- euler + simple scheduler, guidance 3.5, 20–28 steps
Where to place Flux.1 Dev weights on Windows
12B rectified-flow transformer delivering top prompt adherence and typography when loaded via DualCLIPLoader (clip_l + t5xxl) and ae.safetensors.
Place the primary `.safetensors` or `.gguf` weight file inside `models/unet/ or models/checkpoints/` inside your ComfyUI directory, or point extra_model_paths.yaml to your existing shared models drive.
After moving the file, press **R** on the ComfyUI Studio canvas (or click Refresh) so the loader dropdown lists the new checkpoint immediately without restarting the server.
Recommended loader nodes and sampler settings
- **Precision & Memory:** FP16, FP8 e4m3fn, or GGUF Q4_K_S / Q8_0. Cards at the lower end of 12 GB+ (8 GB with GGUF Q4 / FP8) should pair this model with FP8 or GGUF quantization and VAE Decode (Tiled).
- **Sampler & Scheduler:** euler + simple scheduler, guidance 3.5, 20–28 steps. Wire the model output into KSampler and connect the matching CLIP and VAE streams.
- **Installer:** If you have not set up the desktop environment yet, get the Windows build at /download.
Flux.1 Dev Quick Reference for ComfyUI Windows
| Parameter | Recommended Setting | Notes |
|---|---|---|
| Target Directory | models/unet/ or models/checkpoints/ | Refresh canvas with R key after copying |
| VRAM Tier | 12 GB+ (8 GB with GGUF Q4 / FP8) | Use --lowvram or GGUF Q4 on 8 GB cards |
| Weight Format | FP16, FP8 e4m3fn, or GGUF Q4_K_S / Q8_0 | Never load untrusted .ckpt pickle files |
| Sampler Recipe | euler + simple scheduler, guidance 3.5, 20–28 steps | Wire through core KSampler node |
FAQs
Which folder does Flux.1 Dev go into in ComfyUI?
Put Flux.1 Dev inside `models/unet/ or models/checkpoints/` in your ComfyUI installation, then press R on the canvas to refresh node dropdowns.
Can I run Flux.1 Dev on an 8 GB NVIDIA GPU?
Yes — use FP8 or GGUF Q4/Q5 weights via ComfyUI-GGUF and keep tiled VAE decoding enabled so peak VRAM stays under 8 GB.
Run on your hardware. Free forever.
The full ComfyUI engine: open source, fast, extensible, and yours to run however you want. Download the desktop app and open this workflow on your own machine.
