Seedance 2.5 // Hive Watch ads //

Screenshot (2342).png

FULL VIDEO BOTTOM...

Some of you know I'm trying to launch Swiss Hive watch microbrand at the single worst moment in history for both Hive and luxury.
Timing is everything. haha.

Anyways, a few notes from the AI video side of it, since that's where my time is going now in relation to the watches. Trying to figure out how to promote the brand. Do you go classic, do you go with AI. Where is the sweet spot. Is there a sweet spot even?

Seedance 2.5 is arguably the strongest model right now for video generation. It produces genuinely incredible shots. Lifelike enough that some of it passes for documentary footage, some of it reads as properly cinematic. But closeup asset detail is still not ideal. Which is a problem when the asset is a watch and the entire ad is a close-up of the asset. When people think close up, they see "face closeup" as reference for context but assets like this you need to see details of, a human eye never notices. How far away can you really notice a logo or dial details? What about text on dial? You have to be really close.

Can you tell this is AI?

This image above is a frame from the AI generated video. It put in the blisters of the fingers there. It put in a small speck of dust on the thumb. The nail. In the background you can see the finger prints. Insane detail I didnt even prompt.

I talked to @mynewlife about making Hive ads a while back. Hes probably the most talented editor/filmmaker we have on Hive. Its not even "probably" at this point.😅 He saw the potential for AI immediately, but he was also spot on about the limitations. That conversation was months ago, back when the tech was genuinely not nearly as good. It's many times better now. Still not good enough for what I need for low cost but may well be good enough for what Hive needs though. Depending on how we approach things

Cost can be high though. At this stage it's mostly an iteration game, burn generations until something reliable falls out. I do still think we might as Hive be able to get some stuff out of it. Give Raf a few hundred iterations, Im sure he could come up with something as good as this. Me, I stopped at 100 USD spent. lol

But should I have? What if I didnt? What if someone does not stop at first few bad iterations?

We're well past spaghetti eating and seven fingered hands. But detail still slips past the model. My guess: six more months until we're getting Hollywood-level short films out of this.

There are already AI film festivals running. I watched a few at a Monaco conference a month ago.

We're in a parabolic growth that most people havent clocked in on yet. You don't know you're in a parabola when you're in a parabola.