FLUX 3 Image Klein's 5.6-billion-parameter label does not describe the whole system a self-hoster runs. Black Forest Labs' model card says the figure covers the image transformerThe neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.Full definition →; a separate Qwen3-VL text encoder ships alongside, and the card says to always run a prompt upsampler that expands short requests into full scene captions. Source.
On its reference setup, the card reports 31 GB peak GPU memory in BF16 with the upsampler off, plus about 28 GB for its own upsampler. Boxes guide placement and scale rather than acting as a hard mask, so edits can change content outside the requested region.
Terms in this piece · Glossary
transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.
