For media company Black Mixture, AI isn't just a tool - it's an entire pipeline to help bring creative visions to life.Founded in 2010 by Nate and Chriselle Dwarika, Black Mixture takes an innovative approach to using AI in creative workflows. It built its visual reputation in motion design and production using traditional artistry methods and has since expanded its breadth of work to educating artists who want to start incorporating AI tools into their own projects, curating custom AI resources and sharing tutorials on their YouTube channel.
In recent years, Black Mixture has enhanced its creative workflows in apps with GPU-accelerated features. Today, the company has added new AI models and features to its toolkit, running the most complex AI tasks locally on an NVIDIA GeForce RTX 4090-equipped system.
Black Mixture uses generative AI to create or enhance video footage and image assets. This week, RTX AI Garage dives into some of the apps and workflows the company uses, as well as some AI tips and tricks.
Plus, NVIDIA Broadcast version 2.0.2 is now available, with up to a 15% performance boost for GeForce RTX 50 Series and NVIDIA RTX PRO Blackwell GPUs.
An AI for Presentation Iteration is a critical part of the generative AI content creation process, with a typical project requiring the generation of up to hundreds of images.
While models such as ChatGPT and Google Gemini can create a single image fairly quickly, artists like Nate and Chriselle Dwarika require much faster speeds and greater creative control. With on-device AI tools accelerated by RTX, that speed and control is achievable.
Nearly all locally powered generative AI models are accelerated by NVIDIA GeForce RTX. Generating a standard 1024 1024 image in ComfyUI takes 2-3 seconds on my RTX 4090. When you're batch-generating hundreds of assets, that's the difference between an hour, an entire day or more. - Nate Dwarika
Using generative AI, the Black Mixture team can easily draw inspiration from a much wider variety of high-quality images than with manual methods, while refining outputs to their client's preferred vision.
Nate Dwarika uses the node-based interface ComfyUI in his text-to-image workflow. It offers a high degree of flexibility in the arrangement of nodes, with the ability to combine different characteristics of popular generative AI models - all accelerated by his GeForce RTX GPU.
One such generative AI model is FLUX.1-dev, an image generation model from Black Forest Labs. When deployed in ComfyUI on an RTX AI PC, the model taps CUDA accelerations in PyTorch that significantly speeds an artist's workflow. The image below, which would take about two minutes per image on a Mac M3 Ultra, takes less than 12 seconds to generate with a GeForce RTX 4090 desktop GPU.
Made in seconds, not minutes, on RTX. Black Forest Labs also offers FLUX.1 Kontext, a family of models that allow users to start from a reference image or text prompt and guide edits with suggestions, without fine-tuning or complex workflows with multiple ControlNets, extensions that enable more precise control over the image structure, composition and more.
Take the stunning dragon visual below. FLUX.1 Kontext enabled the artist to rapidly iterate on his initial visual, guiding the AI model with additional inputs like poses, edges and depth maps within the same prompt window.
You're scared. I'm not scared. FLUX.1 Kontext is also offered in quantized variants that reduces VRAM requirements so more users can run it locally. It makes use of hardware accelerations in NVIDIA GeForce RTX Ada Generation and Blackwell GPUs.
Together with NVIDIA TensorRT optimizations, these variants provide more than double the performance. There are two levels of quantization: FP8 and FP4. FP8 is accelerated on both GeForce RTX 40 and 50 Series GPUs, while FP4 is accelerated solely on the 50 Series.
Dwarika also uses Stable Diffusion 3.5, which supports FP8 quantization for the GeForce RTX 40 Series - enabling 40% less VRAM usage and nearly doubling image generation speed.
With these visual AI tools, Black Mixture can create photorealistic, original visuals for its clients in a time- and cost-effective way.
Nate and Chriselle Dwarika also use a variety of more traditional content creation tools - all accelerated by RTX. For video content, Nate Dwarika captures footage on OBS Studio using the NVIDIA encoder (NVENC) built into his GeForce RTX 4090 GPU. Because NVENC is a separate encoder, the rest of the GPU can focus on resource-intensive ComfyUI workflows without reducing performance, all while maintaining top encoding quality.
The husband-and-wife duo completes final edits and touch-ups in Adobe Premiere Pro, which includes a wide variety of RTX-accelerated AI features like Media Intelligence, Enhance Speech and Unsharp Mask. They use GPU-accelerated encoding with NVENC to dramatically speed final file exports.
Black Mixture's next project is an Advanced Generative AI Course to help creators get started with using AI in their workflows. It includes a series of live trainings that explore generative AI capabilities, as well as NVIDIA RTX AI technology and optimizations. Attendees can expect 6-8 weeks of live sessions, 30+ hours of self-paced lessons, guest experts, a student showcase, exclusive AI toolkits and more.
Next-Generation Streaming NVIDIA Broadcast version 2.0.2 delivers 15% faster performance for NVIDIA GeForce RTX 50 Series and NVIDIA RTX PRO Blackwell GPUs.
With the performance boost, AI features like Studio Voice and Virtual Key Light can now be used on GeForce RTX 5070 GPUs or higher. Studio Voice enhances the sound of a user's microphone to match that of a high-quality microphone, while Virtual Key Light relights a subject's face to illuminate it as if it were well-lit by two lights.
The Eye Contact
