FLUX.1 [dev]
12B-parameter rectified-flow transformer for text-to-image, guidance-distilled from the FLUX.1 [pro] teacher. Among the most-liked models on Hugging Face (15,986 likes on 2026-11-06). At its 2024 release, BFL said FLUX.1 [dev] surpassed Midjourney v6.0, DALL·E 3 (HD) and SD3-Ultra on prompt following and typography.
Our verdict
A frontier open-weights image model at its 2024 release. BFL's own FLUX.2 [dev] (November 2025) now claims better prompt adherence and typography. Disqualified for anything shipping to paying users by the BFL non-commercial license; pick schnell (Apache-2.0) or pay for the BFL API if revenue is on the line.
Overview
12B-parameter rectified-flow transformer for text-to-image, guidance-distilled from the FLUX.1 [pro] teacher. Among the most-liked models on Hugging Face (15,986 likes on 2026-11-06). At its 2024 release, BFL said FLUX.1 [dev] surpassed Midjourney v6.0, DALL·E 3 (HD) and SD3-Ultra on prompt following and typography.
Strengths
- 12B rectified flow transformer with a hybrid of multimodal and parallel diffusion blocks
- Renders legible in-image text, a long-standing SD weakness
- Strong hands/anatomy compared to SDXL-class models
- Native 1024px, supports up to ~2MP without obvious artifacts
- Mature ecosystem: ComfyUI, diffusers, LoRA training, FP8/NF4/GGUF quantizations
Weaknesses
- Non-commercial license: commercial or production use of the weights needs a BFL license, though the license lets you use generated outputs commercially except as it prohibits
- ~24GB VRAM at BF16; needs FP8/NF4 quantization to fit on 12-16GB consumer cards
- Multi-step sampling: the model card example uses 50 steps at 1024px
- Distillation from [pro] caps the ceiling — guidance can be brittle at high CFG
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| Q4_K_S | 6.8 GB | 9 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of FLUX.1 [dev].
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run FLUX.1 [dev]?
Can I use FLUX.1 [dev] commercially?
Source: huggingface.co/black-forest-labs/FLUX.1-dev
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify FLUX.1 [dev] runs on your specific hardware before committing money.