Black Forest Labs, a global leader in artificial intelligence research, has unveiled a major leap forward in image generation technology. The lab’s latest innovation, the FLUX.1 Kontext model, is transforming how creators, developers, and AI enthusiasts approach digital image creation and editing.

Breaking Down Barriers in Image Generation

Traditionally, artists and developers have relied on a complex mix of models and ControlNets to guide AI-generated images. This often meant juggling multiple tools or employing advanced techniques, such as those found in NVIDIA’s AI Blueprint for 3D-guided image generation, to achieve the desired composition and style.

FLUX.1 Kontext changes the game by offering a unified solution. This model enables both image generation and editing through natural language prompts, eliminating the need for intricate workflows or multiple ControlNets. Users can now guide and refine images with unprecedented ease and precision.

NVIDIA Collaboration: Powering Performance with TensorRT

In a strategic partnership, NVIDIA has worked closely with Black Forest Labs to optimize FLUX.1 Kontext [dev] for NVIDIA RTX GPUs. Leveraging the NVIDIA TensorRT software development kit and advanced quantization techniques, the model now delivers:

• Faster inference speeds
• Lower VRAM requirements
• Smoother, more responsive editing experiences

These enhancements mean creators can iterate quickly and maintain full control over their projects, all from their RTX-powered devices.

FLUX.1 Kontext [dev]: In-Context Image Generation

Launched in May, the FLUX.1 Kontext family accepts both text and image prompts, allowing users to start from a reference image and guide edits using simple, conversational language. This approach removes the need for fine-tuning or complex, multi-step processes.

Key Features

• Character Consistency: Maintain unique traits across various scenes and perspectives.
• Localized Editing: Adjust specific elements without impacting the entire image.
• Style Transfer: Apply the visual style of a reference image to new creations.
• Real-Time Performance: Enjoy low-latency generation for rapid feedback and iteration.

The model’s open-weight design and step-by-step generation process make it accessible and intuitive, supporting both minor refinements and dramatic transformations.

Streamlined Access and Integration

FLUX.1 Kontext [dev] is now available for download on Hugging Face, including TensorRT-accelerated variants. Users can experiment with the model in ComfyUI or the Black Forest Labs Playground, with an NVIDIA NIM microservice version slated for release in August.

Technical Advancements

• Quantization: Model size reduced from 24GB to 12GB (FP8, Ada) and 7GB (FP4, Blackwell), making local deployment more feasible.
• SVDQuant: A new quantization method that preserves image quality while minimizing model size.
• TensorRT Acceleration: Over 2x speedup compared to the original BF16 model running on PyTorch, thanks to optimized use of NVIDIA RTX GPU Tensor Cores.

Getting Started

AI enthusiasts can download the Torch variants of FLUX.1 Kontext [dev] and use them in ComfyUI. An online playground is also available for hands-on experimentation. For developers, NVIDIA is preparing sample code to simplify integration of TensorRT pipelines, with the DemoDiffusion repository set to launch soon.

Expanding the AI Ecosystem: Google’s Gemma 3n and NVIDIA’s G-Assist

The momentum in AI innovation continues beyond image generation. Google recently introduced Gemma 3n, a compact multimodal language model optimized for NVIDIA GeForce RTX GPUs and the Jetson platform. Gemma 3n can be deployed using Ollama and Llama.cpp, supporting popular applications like AnythingLLM and LM Studio.

Performance tests in June 2025 demonstrated Gemma 3n’s efficiency, running with 4 billion active parameters and delivering robust results on RTX hardware.

Meanwhile, NVIDIA’s Plug and Play: Project G-Assist Plug-In Hackathon is underway, inviting developers to build custom G-Assist plug-ins for a chance to win prizes. A dedicated webinar on July 9 will offer insights into Project G-Assist’s capabilities and provide a live Q&A session for participants.