Flux

Flux comes from Black Forest Labs, a team founded by several of the original Stable Diffusion researchers. Its models are 12-billion-parameter rectified-flow transformers, and when Flux.1 launched in 2024 it reset expectations for open image generation — sharper prompt adherence, coherent hands and anatomy, and readable in-image text that earlier open models struggled with. Unlike the SD and SDXL lineage, Flux pairs a large transformer with the T5 text encoder, so it reads long, natural-language descriptions instead of relying on comma-separated Danbooru tags.

The family is really a set of specialized tools rather than one model. Flux.1 Dev is the quality-first workhorse; Schnell is a few-step distillation for near-instant drafts; Krea is tuned for photographic realism and believable skin; and Kontext performs in-context editing, changing an existing image from a text instruction. On Civitai all of them are hosted, so you can jump between Dev, Schnell, Krea, and Kontext without downloading tens of gigabytes of weights for each.

Choose Flux when prompt fidelity, photorealism, typography, or complex multi-subject scenes matter — it is a top open choice for product shots, posters, and text-in-image work. For anime and character art, the SDXL-based Pony and Illustrious ecosystems still lead on style range and sheer LoRA depth. Flux’s own LoRA library is smaller but growing quickly, and because each image runs a 12B model it costs more Buzz per generation than lighter checkpoints — a worthwhile trade when you need the prompt followed exactly.