Flux.1 Schnell
#50 v Text na obrázekBlack Forest Labs · od 2024-08-01 · 2× · naposledy 30. 6. 2026
1
Momentum
FLUX.1 [schnell] is an open-source text-to-image model by Black Forest Labs featuring 12 billion parameters built on a rectified flow transformer architecture. It was trained using latent adversarial diffusion distillation and can generate high-quality images in just 1 to 4 inference steps. Released under the Apache 2.0 license, it is designed for local development and commercial use. Compared to its sibling models FLUX.1 [dev] and [pro], it prioritizes generation speed over maximum image quality.
Vývoj momenta
19.05.17.08.
Vlastnosti
| Fine-Tuning | No native BFL fine-tuning API for Schnell; community LoRA fine-tuning via diffusers/Dreambooth possible |
| Generation Time | 1 to 4 inference steps for high-quality images (distilled for speed) |
| License | Apache 2.0 (free for personal, scientific, and commercial use) |
| Max Resolution | Standard 1024×1024 px; up to 2048×2048 px possible |
| Platform | Hugging Face (weights), GitHub (reference code), ComfyUI, plus API providers such as Replicate, fal.ai, Fireworks, Cloudflare Workers AI |
| Price | Model/weights free (open-source, Apache 2.0); API usage e.g. from $0.0014 per image (third-party provider Fireworks) |
| Price Tier | Free / Open-Source. Weights freely available under Apache 2.0 license – usable for personal, research, and commercial purposes without license fees. |
| Release Date | August 1, 2024 |
| Memory Footprint (GB) | ~24 GB VRAM (BF16/FP16, native); ~16 GB with FP8/Q8 quantization; from ~10 GB with stronger quantization (e.g., NF4). The model additionally requires approx. 24 GB system RAM when loading. |