Generative AI requires massively parallel systems for both training and inference. ChatGPT was trained on 10,000 NVIDIA GPUs for several weeks, but the large language models behind these types of applications are doubling in size and complexity every few months. Applying generative AI to business will require the world’s data centers to become AI factories.
With “Instant Neural Graphics Primitives” (Instant-NGP), the NVIDIA research group demonstrate near-instant training of neural graphics primitives on a single GPU for multiple tasks — gigapixel images, 3D objects, and NeRFs.
In gigapixel image we represent an image by a neural network. SDF learns a signed distance function in 3D space whose zero level-set represents a 2D surface. NeRF uses 2D images and their camera poses to reconstruct a volumetric radiance-and-density field that is visualized using ray marching. Lastly, neural volume learns a denoised radiance and density field directly from a volumetric path tracer.
Approximating an RGB image of resolution 20,000 x 23,466 (469M RGB pixels) with our multiresolution has encoding' with different table sizes. The painting is 'Girl With a Pearl Earring' renovation by Koorosh Orooj
A new iteration moves the experiment from single-prompt outputs into a Weavy node-based workflow. Prompts, references, and style controls are composed as a system rather than a sentence letting Medusa evolve through iteration, art direction, and a more deliberate creative process, and a sharper answer to what authorship looks like alongside AI.
An ongoing study in prompt craft: generating variations of "an image of Medusa" across early AI tools, shifting descriptors, styles, and references with each pass. The work probes our role as creative directors in an AI-mediated landscape — where artistic context, cultural framing, and inclusivity shape whether a generated image carries meaning or simply renders one.
Outpainting with DALLᐧE
