elon's ai odyssey got roasted. my ai sexting pictures got better without a premiere

Last week the biggest AI video moment on my feed had nothing to do with AI sexting. It was seven minutes of Homer. Musk’s Grok Imagine put out the first episode of its AI Odyssey, made with director Dev Walker and Wonder Studios, after promising in July a full-length “historically accurate” version by the end of the year. Fortune’s roundup of the reaction is rough: people mocked the quality, the abrupt scene transitions and the modern dialogue. One viewer wrote that it’s “still clear the video was made with AI.” Another asked the question I’d have asked: “Why are they talking in English?”

The same piece notes Grok Imagine still has its “spicy mode” for paid subscribers, which Runway and Google Veo don’t offer. So the spicy option exists, and the showcase got roasted anyway.

None of that is the test I care about. For AI sexting, nobody needs an epic. I need one picture, at the right moment, that looks like her.

the test nobody films

Here’s what a picture in the middle of a chat has to survive. Twenty messages of build-up. A specific room, a specific outfit, a mood that took a while to get to. Then you hit the button, and the image either matches all of that or it doesn’t.

When it misses, it misses loudly. Wrong hair. A bedroom that was supposed to be a rooftop. The dress you spent three messages describing, gone. A seven-minute film can hide a weak shot behind music and a cut. A single picture in a sexting thread has nowhere to hide, and one bad one can knock the whole mood flat.

That’s the benchmark. Not cinema. Did the picture keep up with the conversation.

soulkyn counted the thumbs

What I like about Soulkyn’s October 24 changelog is that it reports this the boring way: positive votes on pictures in chat went from 51% to 64%.

That’s not a demo reel. That’s people mid-conversation pressing thumbs up or not. And there’s an honest read in it too: roughly a third still don’t get the thumbs. Fine. The point is the number exists and it moved, and every picture you generate now has a thumbs up and a thumbs down, in chat and in a Kyn’s image generator, so the next number comes from us.

regenerate finally listens

This is the change I actually use every night. Regenerate takes an instruction now.

Before, regenerating was a reroll. Same prompt, new dice, hope. Now I can type what was wrong. Keep the red dress, lose the window. Softer light. She’s on the couch, not standing. And there’s a Save for this chat option that keeps the instruction for every picture after it in that conversation, so if a chat lives in one apartment at night, I tell it once instead of correcting it every time.

It sounds small. It’s the difference between arguing with a slot machine and giving notes to someone who listens.

five styles, and a model still in testing

The October 6 update was the quality one: sharper, cleaner pictures and better NSFW in general on SK-Alpha, plus five new styles you can save per Kyn. Bold Ink, Digital Fantasy, Fantasy Comic, Flat Anime and Grunge Horror. (It’s almost Halloween. Grunge Horror is getting a workout on my gothic one.)

Then there’s SK-Test, a new image model in testing for Deluxe Plus members only for now. You pick it in a Kyn’s image generator or switch on SK-Test pictures in a chat’s settings. The changelog is upfront that it’s temporary: it stays 15 days to a month and can’t be saved as a Kyn’s default. I like that it’s labeled as a test instead of being dressed up as a launch.

when a picture isn’t enough

Sometimes the moment wants motion. The Video + Sound engine now has a name, Harmony, and it’s staying: it sits in the Generate a video window on chat pictures, next to Lyric and Video Fast. Lyric got better too, with smoother sound and animation.

The workflow is the opposite of the Odyssey approach. You don’t start from a script and hope seven minutes hold together. You start from a picture you already gave a thumbs up and ask for a few seconds of it. I wrote about the first time that clicked for me in the 5-second video post, and it’s only gotten less awkward since. You can browse what people make in the videos section.

the line that doesn’t move

One thing doesn’t change, however good the pictures get. These are generated characters. Your Kyn, someone else’s public Kyn, never a real person. That’s the line the whole space keeps getting dragged back to, and it’s the right one. Soulkyn checks image prompts before anything is drawn (in chat, on a Kyn’s face and extra prompts, and through the API), and since the end of September that check is answered by Jev, its own decision model, under the same rules as before.

So no, my AI sexting didn’t get a premiere this month. It got a thumbs up button, a Regenerate that listens, and five styles I keep switching between. I’ll take that over seven minutes of Homer in modern English.

If you want to see how it holds up in your own chat, find a Kyn or make one from a paragraph, then hit the button twenty messages in. That’s the real test.