实拍变大片：Seedance 2.0 在 4K 下做 AI 视觉特效完整实战

以下是视频口播的自动转写，可能有错字。

This is real footage, and this is the same shot makes the seed 2.0 in 4K, and there's so much more I can do with it. I can put something behind me that definitely should not be there.

Insane, right? Set my own head on fire, and just keep talking, and change the entire world around me full wide while i'm driving through it, or transform myself.

Right in front of you. And one thing first, pull this up on a big screen. This is the cleanest 4k ai ever pulled, and a small screen does not do it justice.

Hi, i'm adele. I'm not a filmmaker, not a vfx specialist. I just shoot real things and see how far ai tools can push them and everything.

You just saw was created in minutes using cnc 2.0 in 4k inside heixfield mixed with real footage without any prior experience in vfx hours of editing, compositing or render times at a shot, it prompted it and that was it.

And the reason i'm making this video is that getting the effects this clean used to me in hours of work. Students and 4k quality completely changed that you give it your raw clip one sentence and it gives you the output clean enough to drop straight into real work.

That's the real leap and i'm going to show you exactly how you can create the craziest visual effects with already existing footage.

I'm gonna demonstrate three levels of effects. You can create using ai each one bigger and harder than the last level one swap the world around you.

Level two, transform specific elements inside the frame. And level three, the full cinematic showcase, the stuff that looks like it came out of a movie.

This is how you squeeze the most out of aiv effects at the highest quality possible. And one thing before we start, the scale i use for every prompt is free and available in the description.

Grab it and follow along. Now, let's start simple. Level one, swap the world around you. First, we do a location swap with a moving camera.

The easiest place to start. I keep myself exactly as a shot and change everything around me. So here's the clip of me walking and for this, i want to completely change the environment around me.

The workflow here is pretty simple. So learn it once because every shot after this is. The exact same steps. I drop my clip into cloud.

Now, it doesn't watch the video the way you and i do. It reads it as a set of frames like these ones. So, it understands exactly what's in the shot.

Me, my walk, the light, and now it has the full context. Then i turn on the scale, and tell it the one change i want. Swap the background to a desert at sunset.

Right when he snaps his fingers. And just like that, it writes the full prompt for me. And here's what the skill is really doing, the part that makes all of this work.

Based on my request, the skill writes down everything that has to stay exactly as i shot it. My face, my walk, the rings on my hand, every gesture, and the exact handheld camera move, and locks it.

It also writes the exact seconds to change the background. Now, with that prompt, c-dance only swaps the world behind me.

Everything else stays protected. Okay, that's the whole blueprint. My real footage goes in the exact same shot comes back.

Same me, same movement with this insane result in 4k and the desert background. Look at this. But how would you do this?

The old way, you'd green screen yourself, mask out every frame, drop the cutout on a new background and it never matched.

The light was always off. But here only seed ends, re lights me into any scene, any environment. In a matter of minutes, but that was a slow and steady camera movement.

Now, let's speed it up. Same process top down, but talking to the camera while i drive. And here's why driving is worth showing the cameras really moving fast now.

So the new environment can't just sit behind me. It has to move with me at the right speed and change the lighting as it goes.

So i asked for a few worlds at once neon city lava even above the clouds. Look at the prompt, the key list gets bigger here.

It's not just my face anymore. It's the car, the seed build, the rig framing, the whole driving motion, but seed dance should only replace the environment around me.

So let's attach my raw footage, run this prompt in seed dance and see the result. Okay, i mean, look at what it does with the light on its own neon sliding across the car, lava glowing up under my chin.

I didn't grade any of it. Students pulled the lights straight from the generated environment and bounced it onto me. Now, let's move on to the next technique.

Level two, changing an element in the frame. What if you want to change something specific inside the shot? Like adding a creature morphing your own body or literally setting your head on fire.

It's surprisingly easy, but it looks super real. Let me show you how it works. This is my raw footage, and this is what we got.

Now here, the process is pretty much the same. I dropped the clip into the chat and write set my whole head of curls on fire.

It hands me this and again, look how it's already locked my face, my expression, the car, the golden hour light and only added the fire.

Now, think about what setting your head on fire. Normally takes simulating 3d flames for hours, masking your face frame by frame and manually painting the light back onto your skin.

So it doesn't look cheap. That's days of work and a dedicated vfx artist. But here only one prompt and about two minutes.

And one thing to notice, i never masked a single frame students, even spills the orange glow onto my shirt reflection on the car, perfectly matched to the sunset.

This way, my performance stays 100% real. All right. Now, let's move to the next example. This time, let's do a hand morph.

I drop the clip in and ask for one specific thing. Turn my right hand into something unexpected, right when i say the line, transform myself right in front of you.

I want a completely smooth transition, all while i keep talking to the camera, like nothing's happened, till the very end of the clip.

Now, it gives me this prompt. Same deal, it locks my face, my voice, the performance, and the camera. And only touches my right arm.

I dropped that same clip into c-dance with a prompt and run it. The result looks 100% real. No rubber, no obvious cgi, a living thing, doing its own thing, lit to match the daylight on me.

That's the kind of shot people rewind three times looking for the seam spoiler. There isn't one. Okay, let's move on. Let me show you another trick.

That's my boring raw footage to make it interesting. I don't change the background. I add something way more interesting.

Big lizards climbing up the side of the building to make it look real. I give it an input of a lizard made in gpt image 2.0 because when you already know exactly what you want, it's better to show the ai than to describe it.

Then i attach my clip and that reference and tell it exactly what i want. Big lizards climbing up the side of a building.

I didn't want the lizards just stuck behind me. I wanted a reveal. So i asked for a zoom at first tight on them climbing first, then pulling back to show me fully with the right lip sync.

Let's see what we got. I can put something behind me that definitely should not be there. Insane, right? All right, that's how you add something into your shot, plus a camera move you never filmed.

Now, let's move to the one you've been waiting for, and the hardest thing we'll do all video, the handheld cinematic showcase.

The kind of big budget cgi shots that normally need a whole studio. Level 3, full cinematic cgi. This is the hardest one so far on locked off or steady shots.

We've already seen the ai has it easy, the frame barely changes. So we just paints the effects in. But the second cameras in your hand, everything's moving chaotically, the angle, the parallax, the shake, and the effect has to track to all of it without falling apart.

That's the stage fully handheld me in the camera, moving the whole time. Here's the whole trick. I shoot each clip already knowing what i want to see in it.

Then one simple prompt rebuilds the entire world around me. Same clip in a different planet out. I start simple change, my wardrobe and drop it somewhere.

That couldn't exist a wing walker on a flying by plane over floating mountains. I have the prompt and the handheld move.

I shot standing on the ground is now a few thousand feet up and this is the result. I love how the shadows hit my face as the camera pulls back and look at the hair.

She doesn't actually understands physics. It adds a breeze blowing through my hair, even though my original shot didn't have it.

That's the thing that sells it as real and i never touched the mask. Now, let's see the next one. Just me on monkey bars.

I shot it knowing i turn it into a collapsing jungle temple me as the treasure hunter over a bottomless chasm. That's my request and the prompt it gave me straight into c-dance with my raw footage and voila.

And if it feels like this is all happening a little too fast, you're not wrong. It's this quick because the skill does the heavy lifting for you.

The only thing you have to do is be clear about what you want. So far, i've only swap the world around me now. Let's put something alive into it.

Another handheld clip, i asked for three giant long neck dinosaurs towering behind me in a rainy forest, while i turn and react.

This is usually where ai falls apart. Cg creatures look fake and rubbery. But look at this, the camera tracking stays perfectly locked, the background swap looks incredibly real.

And these actually look like living dinosaurs instead of a cheap render. A photo real creature with that much weight is usually the most expensive shot in a movie.

Here, it's just one prompt. And if it can handle a dinosaur, it can definitely handle this next one. This is my absolute favorite shot of the whole video.

I asked for a giant kraken tearing into a ship in a night storm, tentacles everywhere, me right in the middle of it. All actual shot was me walking down a staircase, everything else, the ship, the storm, the whole chaos around me.

Composed it into a single moving shot. This is the prompt it gave me, and this is the result. Looks incredible, right? Now, this is what's actually in these shots, a photo wheel, kraken dinosaurs, towering over trees, and biplane flying over a fictional world in a real studio.

This is the most expensive slowest work. You need creature artists, sculpting, every scale trackers, looking the cg to the camera, move and teams building entire environments from scratch.

That's month of work and a massive budget. I made all four from clips, i shot on a normal camera. All in 4k with no team or a studio and that's what blows my mind.

Tell me what i should transform next in the comments and i'll build the best suggestion in the next video. If you want to try this yourself, the skill is completely free in the description.

Go test out students 2.0 in 4k and see what you can create. And as always, if you found this video helpful, hit the like button, subscribe, and i'll see you guys in the next one.