Blog

Everything at NovaAI is named after space. The models, the characters, the company itself. That’s not a coincidence, but it’s also not the result of some grand naming strategy drafted on day one. The system that exists now emerged over time, project by project, each name chosen for a reason that only made sense as part of a bigger picture in hindsight. Here’s how it came together.

Before NovaAI had a naming system, it didn’t even have NovaAI. The earliest projects were disconnected from each other and named independently. ADA, a home assistant, was a reference to Ada Lovelace. Kyoko, a conversational AI character, got her name from an anime because it sounded right. Neither had a theme behind it, and neither needed one. They were standalone projects with no shared identity, built before there was anything to share an identity with.

When NovaAI became the umbrella, the space theme followed naturally. Nova is a stellar explosion, so everything built under it inherited that language. Model names reflect what they are: a flagship carries the weight of Supernova, something experimental gets the instability of Hypernova, and a large-scale model earns a name like Andromeda. Astra, the reasoning model, takes its name from the Latin word for star. None of this was mapped out in advance; each name was chosen as its project took shape, and a system revealed itself along the way.

The Studio side follows a different convention. Hoshi, the anime image model, is named in Japanese for the same reason the model exists: anime. The characters it generates all carry Hoshi (星, star) as a surname, and their given names are written in kanji chosen to reflect who they are. Each full name reads as something more than the sum of its parts: Aya Hoshi (彩星, colorful star), Yoru Hoshi (夜星, night star), and Mei Hoshi (明星, bright star). Aya was the first character tied to the model, and her name reflects that connection; she represents its creative output. Yoru was originally developed under the internal codename “darkstar,” and the name fit before it was ever formalized. Mei is the newest, and her name mirrors her personality. Like the models, their names are built into the same system that defines everything else at NovaAI.

Kyoko is the exception. She predates the space theme, the kanji conventions, and NovaAI itself. Her name came from an anime and stuck because it worked. No deeper meaning, no system behind it. She’s the oldest project still active, and her name is a reminder that none of this was built with a grand plan in mind.

The naming system at NovaAI is still forming. New models will likely follow the same space-themed conventions, with names chosen to reflect what they do rather than assigned from a list. New characters will carry meaning in their kanji. And somewhere along the way, the pattern will probably extend in directions that don’t exist yet. The best systems aren’t designed. They’re discovered.

-Boof2015

Training Data & Ethics

I train my models exclusively on synthetic data, reinforcement learning, and work that has been permitted to me for AI training purposes. I do not train on unauthorized or copyrighted works. However, I cannot guarantee that the base models I build on top of do not contain unauthorized work, that’s unfortunately beyond my control.

My approach to AI development is guided by respect for human work, ethics, and responsibility. Any similarity to existing art styles or character designs is unintentional and coincidental. If you believe one of my models unintentionally resembles your work or another creator’s work, please email contact@novaml.ai.

I generally view artificial intelligence as a tool rather than a replacement. This directly influences my decisions when making models, and is a primary reason why powerful models like the Hoshi models will always stay closed.

Model Access

Projects are organized into two categories:

• Lab: Research and experimental models. May be open-sourced or API only depending on various factors.
• Studio: Creative projects including characters (Kyoko, Aya) and image generation models (Hoshi, Hoshi-V). These are entirely personal and will never be released to the public.

AI Usage in Official Media

Official Studio media involving Hoshi is created using the Hoshi models. While most content includes human creative direction, editing, compositing, and refinement, some outputs may be fully AI-generated depending on the context. Commissioned work from other creators is rare, but when it occurs, creators are paid fairly for their work. Attribution practices are determined by the creator’s preferences, whether through watermarking, credits, mentions, or other forms of recognition. (or no recognition if requested)

Hoshi-generated content may include an ASCII watermark when appropriate. Watermarked works also contain a unique UUID embedded within the image to prevent unauthorized placement of the watermark.

When commissioned to create work using the Hoshi models (whether 100% AI-generated or not), all works include the watermark and UUID for transparency and recognition.

Hoshi Watermark:

             596666     666664 566666644445527      713222222222221 1966661    566666  444452       
            4000001    800000 50000000000000008  80000000000000006  000004    6000007 000008        
           2000007    000000 5000005    000000  800000             000008    800000  000009         
          30000000000000000 5000003    900000  00000000000000008  0000000000000000  000000          
         3000005    800000 1000002    900000             000000  000009    4000003 800000           
        1000005    900000  0000000000000000  90000000000000008  000008    2000002 000000            
        555553    355555     7544666666999  2888888888888894   255555    7555557 466666

– Boof2015