Phoenix 0.9 is a proprietary foundational image generation model developed by Leonardo.Ai. Introduced as the first model built entirely on the company's internal research and architecture, it represented a shift away from fine-tuning existing open-source frameworks. The model is engineered to prioritize high prompt adherence and is capable of rendering coherent, legible text within images—a task traditionally difficult for many diffusion-based generators.
The model features a specialized Ultra generation mode designed for professional and high-resolution applications. This mode operates with an increased decoder capacity and higher sampling budget, allowing the model to generate images up to approximately 5 megapixels or more. By utilizing a higher-dimensional output regime, the Ultra configuration reduces artifacts in small structures and enhances fine-grain details such as skin textures and intricate patterns.
From an aesthetic standpoint, Phoenix 0.9 is noted for a "warmer" and more cinematic inductive bias compared to its successors. This characteristic makes it effective for creative projects that require atmospheric lighting and soft-focus effects. The model utilizes advanced cross-modal alignment to interpret complex or multi-layered prompts, ensuring that specific objects and spatial relationships described in the text are accurately translated into the visual composition.