Black Forest Labs and NVIDIA Simplify AI Image Editing
Black Forest Labs has announced FLUX.1 Kontext, a new, AI-powered image generation model that represents a watershed moment for users of AI – generate and edit media using natural language and reference images, making it possible to create and edit imagery with unprecedented ease, and blurring the line between artist and developer. Until now, those writers and artists that utilize generative models have been constrained by needing ControlNets (and/or additional tools) to simulate their creativity and curate outputs.
Traditional modalities of image generation require technical steps (depth maps, fine-tuning masks, etc.) to create artistic output – this can be unnecessarily complicated and time consuming. FLUX.1 Kontext eliminates that complication and saves the user time; the creator describes in words what they want the model to output, or provides it with a starting image. From there, the model interprets the user’s language step-by step, producing coherent and precise edits by remaining very close to the intent of the user.
What can’t be understated with FLUX.1 Kontext is that the model supports multi-turn editing; the user can in fact make changes incrementally and still be confident that key elements will be consistent (like style, structure or identifiable character). From changing colours to re-conceptualizing entire scenes, FLUX.1 Kontext preserves the original visual idea.
“We are changing how people create with AI,” offered a representative from Black Forest Labs. “Now, you do not need five separate tools to do one thing; it’s just one model to do what you want, and your words.”
NVIDIA collaborated with Black Forest Labs to enhance the performance of the FLUX.1 Kontext。。Together they were able to optimize the model on RTX GPUs with TensorRT and quantization, reducing generation time and VRAM usage. The quantized models can now run locally on most GeForce RTX systems and have even reduced the model size from 24GB to 7GB in some cases, without any projected loss of quality in the images generated.
“TensorRT has more than 2x speed improvement compared to the baseline PyTorch versions”, NVIDIA’s technical team stated. “With FLUX.1 Kontext [dev] we can help creators do more with no need for expensive setups or complicated software stacks.”
The model is currently available through ComfyUI and Black Forest Labs Playground, with a version coming this August for NVIDIA NIM micro-services.
Notable features include character consistency across scenes, localized editing of elements; real-time responsiveness; and style transfer with seamless consistency.
Overall, this partnership will help democratize image generation and editing and become more accessible intuitive and fast for everyone from designers to developers to everyday creatives.
____________________________________________________________________________________________________
Latest Stories
Houghton Mifflin Harcourt (HMH) Unveils Gen AI Suite
POST Luxembourg Partners with Ericsson to Boost Network Security
