How I made "Objection" (the full breakdown)

Y'all asked how I put these AI dramas together, so here's everything from one of my latest, "Objection." It's 98 seconds, vertical 9:16, at an Atlanta garden wedding at golden hour. The best man, Jordan, stands up and objects. It isn't about love, though. He's exposing that the groom, Marcus, owed a $40K loan and hid his money problems. Simone hands back the ring and says, "You already made a decision for me. Now I'm making one for myself." Then she walks away down the aisle.

Step 1: Script before anything else

I wrote the whole story out scene by scene before I made a single image. The hook has to land in the first 2 to 3 seconds, so I opened right on the objection with no slow setup. I ended up with 11 scenes, each with one location, who's in it, and one line or one action. If a line doesn't move the story, I cut it, because every extra line is another clip you have to generate and fix.

Step 2: Lock your cast first

This is the step most people skip, and it's why their characters look different in every shot. I made a reference portrait for each of my 4 characters and wrote down exactly what they look like:

Simone: deep brown skin, low bun, ivory lace off-shoulder gown, cathedral veil
Marcus: fade, beard, black tux, bow tie, white rose boutonniere
Jordan: dark skin, low cut, black tux, silver tie
Officiant: in his 60s, gray goatee, black robe

I used those same portraits and descriptions for every single scene. Same outfit, same hair, same details, every time.

Step 3: One image per scene

I made 11 images, one per scene, always building from the reference portraits. No collages or grids, just one clean shot each. If a face or outfit drifted from the reference, I threw it out and made it again. Also watch where people are standing and looking, because blocking matters for continuity.

Step 4: Animate each scene

I turned each image into a clip of about 11 seconds with built-in dialogue audio. Two rules made a big difference:

Only one person talks per clip, so the voices don't overlap or get confused.
Keep the action simple, like a step forward, a turn or a look. Skip the extra hand gestures, because that's where AI movement starts looking off.

Step 5: Check every line

I ran every clip's dialogue through speech-to-text and compared it to my script. Sometimes a clip sounds fine at first but has a wrong word or a mumble. Any clip that didn't match got regenerated.

Step 6: Stitch and trim

I put the clips together in order and cut the dead air at the start and end of each one. That took it from 120 seconds down to 98.5. Tighter is better.

Step 7: Export for phones

My first export was 34 MB and wouldn't even open on a phone. I re-exported at 720p to get it under 10 MB, and it played fine. Always test your file on your own phone before you post.

What I'd fix next time

I'm keeping it real with y'all. Jordan's step forward came a beat late, a side-eye glance from Marcus crept into one shot, and the final silent scene of Simone walking away felt a little too quiet. Next time I'd tighten that timing, regenerate that shot, and add sound under the ending.

3 takeaways you can use today