Flux Kontext Pro

FLUX.1 Kontext is a flow matching model, a rectified flow transformer that works in the latent space of an image autoencoder rather than on raw pixels. According to Black Forest Labs' technical report, it is trained with a single velocity prediction objective on a concatenated sequence of context and instruction tokens, which lets one unified model handle both text-to-image generation and image editing without separate pipelines.

The big change from earlier FLUX models is in-context generation: you can prompt with an image plus a text instruction, not just text. That enables local edits, scene transformations and multi-step refinements while preserving a reference character or object across different scenes and viewpoints. Black Forest Labs says the Pro version is built for iterative workflows and runs up to 8x faster than leading diffusion-based editors at the time of release.

Black Forest Labs openly lists the model's limits: excessive multi-turn editing sessions can introduce visual artifacts and identity drift, instruction following fails in rare cases, and the model has limited world knowledge. Because Kontext regenerates the full image on each edit rather than only a masked region, sharpness and fine detail can degrade over long editing chains.

On safety and provenance, the FLUX API attaches cryptographically signed C2PA metadata to outputs to mark them as model-generated. Prompts, uploaded images and outputs pass through content filters, including non-adjustable CSAM and NCII filters from Hive, and Black Forest Labs worked with the Internet Watch Foundation to filter training data before release.