AI Images, LLM, ChatGPT

Read in: IT 🇮🇹   EN 🇺🇸

During the Varese RetroComputing 2024, a fair increasingly rich in not only vintage but also current content, I had the pleasure of interviewing Michele Petino, who professionally deals with 3D modeling, rendering, virtual tours and lessons on the implementation of AI in the production of images and renders. It was a meeting full of technical ideas, where we talked about both AI images and the world of three-dimensional modeling, passing through advanced tools such as ComfyUI, FLUX.1 [dev], Civitai, Stable Diffusion, ChatGPT, SketchUp and D5 Render.

Modular AI workflow

Michele uses Intel Core i9 and NVIDIA GeForce RTX 4090 hardware with ComfyUI, Stable Diffusion, and FLUX Dev.1 to produce AI images in a modular, customized ecosystem.

ComfyUI: Nodal Interface

ComfyUI is a node interface that allows you to build advanced pipelines for Stable Diffusion via visual charts, connecting operations such as model loading, prompt editing, and filters in a single workspace.

Stable Diffusion: generation engine

Stable Diffusion is an open-source text-to-image model based on diffusion and denoising processes, capable of transforming textual descriptions into high-resolution images.

Workflow Civitai by Tenofas

The Civitai workflow used by Michele was developed by Tenofas and integrates the Flux Dev.1 model with LLM nodes for the automatic generation of prompts directly in ComfyUI (Tenofas' modular FLUX Workflow v.4.3 – OpenArt, Looking for the best workflows/methods to get amazing images …).

This modular solution supports txt2img, img2img, FaceDetailer applications, and upscaling capabilities, providing an "all-in-one" experience for the end user.

The version tested at the fair was downloadable on Civitai at the dedicated link, with a simplified GUI to load and launch entire sequences of nodes in one click.

FLUX Dev.1: model and integration

FLUX.1 [dev] is a 12 billion parameter rectified flow transformer for generating images from text, created by Black Forest Labs for non-commercial applications.

Thanks to API clients, it can run in ComfyUI as a dedicated node, allowing you to easily toggle between Stable Diffusion, Flux, and other diffusion backends in the same pipeline.

Prompting with LLM and ChatGPT

LLMVISION or PromptGenerator nodes allow you to integrate language models (GPT-4, Claude) into the ComfyUI flow, generating descriptive and complex prompts without leaving the interface.

Alternatively, ChatGPT can be used externally to come up with prompt ideas, then manually importing them into ComfyUI or the Tenofas workflow for quick and detailed iterations.

Photorealistic 3D Modeling and Rendering: SketchUp and D5 Render

In the second part of the interview, Michele takes us into his professional field: the creation of 3D animations and architectural renderings. Artificial intelligence is less present here, but the technical level remains high.

To model the environments, Michele uses SketchUp, one of the most popular software for 3D design in the architectural field. With SketchUp, he builds volumes, structures and furnishings, while maintaining a high level of precision and compatibility.

The final rendering is done via D5 Render, a real-time rendering engine that combines visual quality and speed. D5 Render also uses some AI technologies for the processing of lights, shadows and materials, generating photorealistic results in a much shorter time than traditional software.

Thanks to D5 Render, Michele is able to produce animations, walkthroughs and images of the highest quality, often used in promotional or architectural fields.

During the interview, Michele showed us an animation of an airplane.

Subscribe to my channel and stay up to date

To stay up to date on topics related to computers and vintage electronics, but also more modern, I invite you to subscribe to the YouTube channel @ValorosoIT and activate the notification bell so you don't miss the next videos. I'm also present on TikTok, Instagram, Facebook, Threads ...