Flux

FLUX is the image generation model from Black Forest Labs, founded by several of the researchers behind Stable Diffusion. Released in mid-2024, it immediately established itself as the state of the art on prompt adherence and overall generation quality, surpassing Stable Diffusion XL and matching or beating Midjourney on several benchmark categories.

The defining strength of FLUX is how accurately it follows complex prompts. Describe a scene with many specific elements — the lighting, the objects present, their spatial relationships, the camera angle, the mood — and FLUX renders it with a fidelity that earlier models struggled with. Prompts like “a woman in a red raincoat standing at a crosswalk in Tokyo, rain-slicked streets, neon reflections, 35mm film grain” produce outputs that feel like they were art directed, not hallucinated.

Text rendering is another genuine breakthrough. Most image generation models produce unreadable, misspelled, or distorted text when asked to include words in an image — a limitation that makes them impractical for marketing graphics, posters, and social media content. FLUX renders legible, correctly spelled text in images with enough reliability that it’s usable for real production work.

Three variants are available: FLUX.1 [schnell] is the fastest model, open-weight, suitable for high-throughput applications. FLUX.1 [dev] is the open-weight research model with higher quality, available for download and local use under a non-commercial license. FLUX.1 [pro] is the highest-quality version, available via the Black Forest Labs API and through integrations in Replicate, fal.ai, and ComfyUI cloud services.

For developers building image generation into applications, the FLUX API is the most direct path. For local creative use, FLUX.1 [dev] runs on consumer GPUs (12GB+ VRAM recommended) with ComfyUI or similar frontends.