Overview
- The FLUX.2 family debuts across managed endpoints (pro and flex) and an open‑weight dev model, with weights on Hugging Face and access via the BFL Playground, BFL API, and partners including Cloudflare Workers AI.
- New capabilities emphasize photorealism, multi‑reference editing for consistent subjects and styles, direct pose control, cleaner text rendering, and image editing or generation up to 4 megapixels.
- The flagship FLUX.2 [dev] checkpoint is a 32‑billion‑parameter model that initially required about 90GB of VRAM, or roughly 64GB in lowVRAM mode.
- NVIDIA and ComfyUI optimizations introduce FP8 quantization and RTX performance tuning that reduce VRAM needs by around 40% and improve runtime by roughly 40%, with ComfyUI weight streaming enabling system RAM offload at some performance cost.
- Black Forest Labs frames FLUX.2 as rivaling closed models for quality and control, with those claims presented as company benchmarks pending broader independent validation.