One character staying consistent is already hard. So I gave Wan 3.0 three completely different characters and tried something harder: putting them into the same sequence without making them feel randomly generated. Built in Magnific. Here’s what happened 👇

These were my starting points. Three characters. Three different outfits. Three different visual identities. I wanted each one to remain recognizable once the scene started moving, instead of slowly turning into the same generic AI person.

This is where AI video usually breaks for me. A character looks perfect in one shot, then the camera cuts and suddenly the face, clothes or proportions have changed. With multiple characters, that problem gets multiplied. That’s why I wanted to stress-test Wan 3.0 this way.

Here’s the actual process inside Magnific. What I like about working from references is that I’m not asking the model to make every creative decision for me. I define who belongs in the scene first. Then Wan 3.0 has to make them perform.

For me, this is the difference between generating clips and actually building scenes. If characters can survive cuts, movement and interactions while keeping their identity, you can start thinking beyond one-shot demos. That’s the part I’m watching closely. Wan 3.0 on Magnific:

That’s awesome

The consistency here is getting seriously impressive

Character consistency is improving so fast. Wan 3.0 handled this really well !

Awesome sharing

Thanks for sharing

Really useful information here.

Excellent share

This feels way closer to filmmaking than prompting at this point

Awesome share

This is a gem. 💎

Keeping three distinct characters consistent in the same sequence is seriously impressive.

The character consistency is what stands out most to me

Amazing one

Three characters, one sequence, and consistent storytelling—that’s where AI gets really impressive.

Wow, it's amazing

Outstanding work with magnific !

So impassive share