Mark Theriault founded Fity, a startup that imagines a line of clever cooling products. A cold drink holder with a freezeable pack to keep your drink cold for a long time without ice mess. The entrepreneur built one unit at a time, starting with printing 3D products in the basement and before eventually expanding into mass production.
Establishing a consumer product company from scratch was a high order for one person. Moving from preliminary sketches to creating designs was a major challenge. To realize his creative vision, Theriault relied on AI and his Nvidia GeForce RTX equipment system. For him, AI is not just a tool, but the whole pipeline that helps him achieve his goals. See below for more information about his workflow.
Plus, the GeForce RTX 5050 laptop arrives today at retailers around the world starting at $999. GeForce RTX 5050 Laptop GPU Functions 2,560 NVIDIA BLACKWELL CUDA CORES, 5th Generation AI Tensor Core, 4th Generation RT Core, 9th Generation NVENC Encoder, 6th Generation NVDEC Decoder.
Plus, Nvidia's Plug and Play: Project G-Assist Plug-In Hackathon – running effectively until Wednesday, July 16th. Invite developers to explore AI and build custom G-assist plugins to get a chance to win prizes. Save the dates for the G-Assist plugin webinar from 10am to 11am on Wednesday, July 9th to learn more about the features and fundamentals of Project G-Assist and participate in a live Q&A session.
From concept to completion
From sketching to computer-aided design to rapid prototyping, until Theriault Tinkers, with a potential fish flex cooler design with traditional methods, finding the right vision. A unique aspect of Fity Flex design is that it can be customized with the appeal of fun and popular shoes.
For packaging design inspiration, Theriault uses his preferred text-to-image generation AI model, stable diffusion XL, for prototypes, which runs 60% faster with the NVIDIA Tensort software development kit, using a modular node-based interface Comfyui.
COMFYUI granularly controls the user through every step of the generation process, including prompting, sampling, model loading, image conditioning, and post-processing. This is perfect for advanced users like Theriault who want to customize how images are generated.
Based on the Nvidia Blackwell architecture, Nvidia and GeForce RTX GPUs include a fifth-generation tensor core designed to accelerate AI and deep learning workloads. These GPUs work with Pytorch's CUDA optimization to seamlessly accelerate Comfyui and reduce flux generation time.
Comfyui can also add control nets (AI models that help control image generation) that Theriault will guide human poses, set configurations via depth mapping, and convert doodles to images, etc.
Theriault creates his own fine-tuned models to keep his style consistent. He used a low-rank adaptation (LORA) model (small and efficient adapters are included in a particular layer of the network) to allow for hyper-customized generation with minimal computational cost.
“Over the last few months I have been using a custom flux rora I trained at home to completely AI-generated product images. My RTX 4080 Super GPU is essential for getting performance that needs to be trained and repeated quickly.” – Mark Theriault, Founder of Fity
Theriault also uses the generated AI to create marketing assets such as Fity Flex product packages. He uses Flux.1. It is excellent at generating easy-to-read text within images, addressing the general challenges of image models from text.
Flux.1 models can normally consume more than 23GB of VRAM, but Nvidia worked with Black Forest Lab to help reduce the size of these models using quantization. The model is then accelerated with Tensort, offering up to twice the speedup over Pytorch.
To simplify using these models in Comfyui, Nvidia created the Flux.1 Nim Microservice. This is a containerized version of Flux.1 that is loaded with Comfyui and allows FP4 quantization and Tensort support to be enabled. When combined, the model will be VRAM over 11GB, improving performance by 2.5x.
Theriault uses the Blender Cycles app to render the final file. For 3D workflows, NVIDIA offers AI blueprints for 3D guided generation AI, which makes it easy to position and configure 3D images, so anyone interested in this method can start right away.
Finally, Theriault uses a large-scale language model to generate marketing copies tailored to search engine optimization, tone and storytelling, and complete tasks that typically cost thousands of dollars in legal costs and considerable time to complete.
“As a single band that generates a lot of content, having on-the-fly generation capabilities for my product design really helps speed things up.” – Mark Theriault, founder of Fity
All the textures, every word, every photo, every accessory was micro-decided, Theriault said. He added that AI helped survived “Death by a Thousand Cuts” which could stall the founders of solo startups, he added.
Every week, RTX AI Garage The blog series features community-driven AI innovation and content for those looking to learn more about NVIDIA NIM microservices and AI blueprints, as well as buildings AI Agentcreative workflows, digital humans, productivity apps, and more.
Connect to nvidia ai pc on Facebook, Instagram, Tiktok and x – And you will be notified by subscribing to RTX AI PC Newsletter.
Follow Nvidia Workstation LinkedIn and x.
look Let me know Regarding software product information.
