FLUX.1 Kontext
In-context image generation and editing — multi-turn edits with character consistency.
About
FLUX.1 Kontext is a suite of generative flow-matching models from Black Forest Labs, released May 29, 2025, that unifies image generation and editing in a single architecture. Unlike traditional text-to-image models, Kontext accepts both text and image inputs: generate from scratch, then iteratively retouch with plain-language instructions while it preserves characters, style, and composition across turns. The paper (arXiv:2506.15742) reports state-of-the-art multi-turn consistency at 3–5 second inference for 1MP images — up to 8x faster than leading alternatives. It ships as Kontext [pro], [max], and an open [dev] variant.
The honest picture: it's the strongest option for iterative image editing with open-weights availability — [dev] runs locally via ComfyUI or as an NVIDIA NIM microservice (quantized to 7–12GB for RTX cards). But [dev] is non-commercial licensed, there's no explicit mask support (editing is text-driven), and for pure text-to-image the base FLUX models remain competitive. Adobe integrated Kontext [pro] into Photoshop Beta's Generative Fill in September 2025.
What it can do
- Unified text-to-image and image-to-image editing in one model
- Multi-turn editing with character and style consistency
- Local edits, scene transforms, and style reference
- 3–5s inference at 1MP resolution
- KontextBench: 1,026 image-prompt pairs across 5 task categories
- Photoshop Beta Generative Fill and NVIDIA NIM integrations
Setup
Use via API partners (Replicate, fal) or run FLUX.1-Kontext-dev locally: accept the license on Hugging Face, then load in ComfyUI or deploy the NVIDIA NIM container.
Prerequisites
For local use: 12GB+ VRAM (FP8) or an RTX 50-series card (7GB FP4); a Hugging Face account.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 46; Trending score 85/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.