A New Reason to Back the Byron AI

The last three weeks have significantly underlined the importance of supporting the combined efforts of Castalia and Infogalactic to develop a custom AI system that will protect and enhance the ability to produce fiction without limits being imposed by the major AI providers. A number of things have happened since we launched the Byron AI crowdfunding campaign, including:

• The continued degradation of Claude’s ability to creatively hallucinate.

• The expanded interference with subject matter: one news site was aggressively denied the ability to use a major AI system to write about Hayden Panattiere’s death due to sensitivity concerns about its recent occurrence in its article announcing Hayden Panattiere’s death.

• Deepseek increasing its token price by 4.5x due to its upcoming IPO.

• OpenAI actively blocking ChatGPT users from being able to intentionally imitate the style of authors both living and dead.

Even though the campaign hasn’t ended yet, the dev team has already made solid progress on a number of peripheral items, including the ability to accurately imitate various literary styles and eliminate the various AI tells, including these abominations.

The eliminations of these tells have been made with tactical precision. And that matters. Let the weight of that sit for a moment.

To say nothing of its unbreakable habit of naming every third character either Chen, Martinez, Okwonko, or Chen (no relation).

Now, here’s where it gets exciting. The success of the campaign has meant that we’re going to be able to acquire the more serious hardware that we had been eyeing for hosting the Byron AI system. At the same time, the release of a new video engine, MiniMax H3, turns out to be the best engine for producing lip-synced video, because it is not frame-based like the two best predecessors, Clarify and Avata, but character-based. This permits movement during speech and speech from various positions.

Now, I’ve been using music videos to explore the technology, because watching a music video is interesting for three minutes whereas a drama, comedy, or romance that only lasts three minutes barely serves to grasp one’s attention. But the requirements and the functionalities are essentially identical. So you can probably understand the direct connection between a) being able to produce fictional text and b) being able to bring that fictional text to visual life.

Here is an example of frame-based lip-syncing from the recent Silenziosa song. There is an amount of animation, but it’s all based on the single frame from which it was built.

Next is an example of character-based lip-syncing. This was just the first test, so we kept the challenge very simple and the movement range very limited, since if Minimax H3 couldn’t handle it, then it wouldn’t be able to handle more complicated challenges. While you can see the current weakness of character-based animations; lacking a profile image for the character, the visual accuracy is a little weaker. But that’s a known problem which we know how to fix; the important thing is that it worked! Because the ability to have spoken dialogue in motion is absolutely vital for non-music visual entertainment.

So we then threw a much more difficult test at it by not isolating the vocal track, utilizing a profile shot, then adding a turn, all in rapid sequence. The test was entirely successful; the animation ending before the final word in the audio track is an artifact of the Substack upload and the video plays properly in its entirety in its original form.

If you’re wondering where that particular track came from, an amount of old Psykosonik audio and video recently surfaced, including long-lost demo tracks and concert footage. I’ve been given permission from my bandmates to experiment with what is possible via AI; here is a brief clip from our second show at Glam Slam, featuring a song we never released publicly. In case you don’t recognize the drummer, by the way, that is the longtime Audio Director at Epic Megagames, responsible for the sounds and music for some of the biggest games over the last 20 years.

So what does any of this music and video stuff have to do with the Byron AI campaign? The answer is this: the hardware that is necessary for the production of AI text is very well-suited to run this particular video engine. So supporting our ability to do the one will simultaneously support our ability to do the other. In other words, this isn’t just a first step toward independent book production of the sort you’ve already seen with Dorian Vane and the Episodios Nacionales, it’s also a first step toward independent animated movie production.

We’ve even added a special new reward to the campaign, as you can request a music video for the song of your choice by Soulsigma, Vibe Patrol, or Psykosonik. And don’t forget, our number one backer has agreed to double the total amount of the campaign, so you’re getting 2x bang for your support buck.

And speaking of teaming up with Lord Byron, here is a song in which the pitch for the Byron AI campaign was modified and transformed into a neo-Byronic by Soulsigma called Last to Know.

Do not ask me what is love,

No, never look into my soul!

For if anything is there

Then I will be… the last to know.

Back the Byron AI