I'm watching this video for the third time and still cracking up. The guy is waging war on his couch with a metal camera tripod. Eight seconds later - he's in Vietnam.
Seriously. Sits on the floor of his room, couch and a guitar on the wall behind him, aiming the tripod somewhere in the direction of the TV.
AI turns him into a special-ops soldier by a campfire in night jungles. Same pose. Same head tilt. The tripod becomes an M4, the guitar becomes the darkness of a tropical night.
Then he lies down on a blanket in the middle of the room and aims while prone. AI turns him into a sniper in the mud under pouring rain. Pose identical down to the finger on the trigger. Which doesn't exist. It's a tripod.
Third shot - comes out of the bathroom with the tripod at the ready. AI turns him into a soldier rising from a stream in the jungle. What was in the bathroom is not specified.
And here's the thing to understand:
AI no longer needs a perfect source. It needs a pose. Sloppily acted, in sweatpants, with a tripod, a mop, or an iron. Pose transfer works so precisely that a room with a guitar becomes a combat location.
Framework:
→ RECORD - act the pose with whatever is at hand
→ REIMAGINE - ChatGPT changes the world
→ ANIMATE - Seedance brings it to life
Between an actor on a paycheck and a guy with a tripod, there's only one difference left: the guy doesn't have an agent.
This week I'm breaking down a couple more finds like this.
Five days ago he was fighting in Vietnam alone. Today he's in World War II with a friend.
Same guy. Same look, only now there's two of them. The friend is in a red t-shirt, both walking through a residential block in Indonesia with airsoft.
Seventeen seconds later both are in 1944 uniforms, in helmets, on a ruined battlefield, the neighbor's house wall riddled with bullets.
But that's not the point. The point is that two AIs are already fighting for the right to film them.
The same source scene was run through Seedream 5.0 Lite and Dreamina Seedance 2.0. Both bring back World War II, but differently. At 0:06 Seedance adds a burning pillar on the left side of the frame, Seedream leaves clean mud without fire. At 0:12 by the ruined wall Seedance gives burning debris and a close dark shot. Seedream holds both heroes in the frame with a bit more light.
And here a new question shows up, one that didn't exist before.
Before, people picked an AI by whether it worked or not. Now they pick by directorial style. Seedance is closer to Tarantino's camera work - fire and close faces, darker. Seedream is closer to Malick's - both in frame with even light, stable camera.
Not long ago the main question about AI video was "can it". Today the main question is "which of the two". This guy from Indonesia already has answers for both buttons.
Scheme:
→ STAGE - play the scene keeping the face and gesture
→ FORK - run it through two tools in parallel
→ PICK - choose whose style matches your vision
AIs now have directorial styles. Zero before, two now, five soon. Hollywood never had that many.
This week I'm breaking down a couple more finds you'll like.
I'm watching this video for the third time and still cracking up. The guy is waging war on his couch with a metal camera tripod. Eight seconds later - he's in Vietnam.
Seriously. Sits on the floor of his room, couch and a guitar on the wall behind him, aiming the tripod somewhere in the direction of the TV.
AI turns him into a special-ops soldier by a campfire in night jungles. Same pose. Same head tilt. The tripod becomes an M4, the guitar becomes the darkness of a tropical night.
Then he lies down on a blanket in the middle of the room and aims while prone. AI turns him into a sniper in the mud under pouring rain. Pose identical down to the finger on the trigger. Which doesn't exist. It's a tripod.
Third shot - comes out of the bathroom with the tripod at the ready. AI turns him into a soldier rising from a stream in the jungle. What was in the bathroom is not specified.
And here's the thing to understand:
AI no longer needs a perfect source. It needs a pose. Sloppily acted, in sweatpants, with a tripod, a mop, or an iron. Pose transfer works so precisely that a room with a guitar becomes a combat location.
Framework:
→ RECORD - act the pose with whatever is at hand
→ REIMAGINE - ChatGPT changes the world
→ ANIMATE - Seedance brings it to life
Between an actor on a paycheck and a guy with a tripod, there's only one difference left: the guy doesn't have an agent.
This week I'm breaking down a couple more finds like this.
A guy drowned in his own room. The neighbors heard laughter. AI heard panic.
Eight seconds of video, four transformation scenarios, one actor in an orange shirt.
- The guy raises a finger and smiles.
- At 0:01 same pose - AI floods the room with rain, water running down the shirt.
- At 0:03 same pose - AI sets the studio on fire, soot on his face, flames on every side.
- At 0:04 same pose - AI covers everything in frost, the shirt frozen over, a smile through the ice.
- At 0:05 the guy puffs up his cheeks and holds his breath - AI builds an ocean at twenty meters deep, hair floating, bubbles around his hand.
One actor. One room. Four elements.
The shift is this.
AI used to change the world around a given pose. Now AI reads the emotion off the face and builds the world to match. Hold your breath - you get an ocean. Smile through squinted eyes - you get snow.
A Hollywood underwater scene needs a stunt double in a wetsuit and a week of prep. Here - puffed cheeks, two seconds.
Scheme:
→ FACE - show the right expression (holding breath, smiling through cold)
→ READ - AI reads the emotion as a cue for the environment
→ BUILD - AI builds the world around that emotion
Holding your breath is worth more now than knowing how to swim.
This week I'm breaking down a few more finds. One of them breaks everything I said above.
Marvel and Game of Thrones casting is closed. The guy with a houseplant won.
Two transformation scenarios, same actor, same ordinary room.
At 0:00 he's petting a potted plant - AI turns him into a northerner in furs on snowy cliffs, the white direwolf Ghost under his hand. At 0:05 he points a flashlight around a dark living room - AI turns it into a lab, a Venom-style symbiote breaking through the glass. At 0:10 he stretches his arms wide in the same room - AI turns him into a Marvel hero on a hilltop during a storm, rocks levitating around him.
One actor. One living room. Three different franchises.
Now look closer.
What you filmed yesterday on your phone is no longer one video. It's footage that fits any genre you haven't picked yet. The genre gets picked after the shoot, not before.
A studio shoot day costs as much as a new car. Every frame is shot for one specific film. If the film doesn't release - the footage sits in the archive. It's raw material for five different projects at once.
Scheme:
→ FRAME - shoot any gesture in any room
→ CAST - pick the genre (Marvel, GoT, horror, sci-fi)
→ RENDER - AI assembles the final frame in that genre
Every frame you shoot today is a contract with films that don't exist yet.
Next post - where this method breaks. Not on the genre. Elsewhere.