Microcosmo v2 Mixed

Microcosmosv2 is a generative image model based on the Stable Diffusion 1.5 architecture, designed to offer a high-quality, Brazilian-developed solution for generative AI. As the successor to MicrocosmosCC0, it expands its capabilities with anonymized data collected from public sources, compliant with LGPD (Brazilian data protection law) and copyright regulations.

The National Data Protection Authority (ANPD) in Brazil has emphasized that personal data used for AI training must be anonymized, and such data should not be retained after the training process.

The automated use of works, such as extraction, reproduction, storage, and transformation, in data and text mining processes in artificial intelligence systems, does not constitute a copyright infringement when conducted by organizations and institutions engaged in research, journalism, museums, archives, and libraries, provided that:

I – it does not aim for the mere reproduction, display, or dissemination of the original work itself;
II – the use occurs only to the extent necessary to achieve the intended purpose;
III – it does not unjustifiably harm the economic interests of the rights holders;
IV – it does not compete with the normal exploitation of the works.

• Personal Data Protection: The model complies with LGPD by anonymizing all personal data and ensuring that no data is stored after training, respecting privacy and informational self-determination.

📂 Training Data

• Sources: ArtStation, Twitter, Discord, and other public social media platforms.

• Process: Anonymized data used solely for training, with no post-training storage.

• Purpose: Pattern extraction without directly reproducing original works. The training data also contains AI-generated content, ensuring a diverse range of examples for improved model performance.

📜 Terms of Use

• Commercial Use: Allowed only for works incorporating significant human editing. The results generated will be considered copyright-free in themselves and may be used commercially, only if these results are incorporated into other works or if they are substantially processed or edited by a human being, considered as director or editor of the work. The director or editor who carries out substantial editing of the work may obtain rights to the complete work, but not to individual elements (such as generated images, patterns or specific texts). The objective is to avoid the appropriation of individual rights over components generated by AI, ensuring that the final work is the result of human creativity.

• Ethical Practice: Intended for creators who value transparency, privacy, and responsible innovation.

  ---

According to the National Council for Scientific and Technological Development (CNPq), certain types of creations cannot be registered or protected as intellectual property. These include abstract ideas, methods, plans, concepts, scientific discoveries, mathematical theories, and algorithms that lack concrete and practical applications. “The principles guiding these exclusions are rooted in the necessity of keeping foundational knowledge and generic expressions accessible for the advancement of science and technology, rather than subject to monopolization.” In the context of artificial intelligence, these principles are particularly relevant. AI-generated content and model weights often fall within the scope of creations excluded from intellectual property protection. Model weights, for instance, represent a set of numerical parameters derived from mathematical optimization during the training process. These weights are essentially computational artifacts, devoid of human creativity or distinctiveness, and therefore do not meet the originality or authorship criteria required for intellectual property registration. Similarly, AI-generated content—such as images, text, or other media—may also fail to qualify, if it lacks significant human involvement or does not demonstrate sufficient originality.