How to Make a Wedding Love Story Using AI (Full Tutorial) — Transcript
Full transcript
- 0:00A love story I made for my friend's
- 0:01wedding went viral, getting around 500
- 0:03million views across social media.
- 0:06Agencies charge thousands for films like
- 0:08this. Today, I'm showing you how to make
- 0:10one yourself. Just steal my prompts.
- 0:12[music] Hi, I'm Adil. I generated the
- 0:14whole thing with Claw and Cedence 2.5 in
- 0:16Geeksville. Today, [music] you're
- 0:18getting the full workflow. How to fix
- 0:19sloppy AI faces, nail the scale, and
- 0:22make the film look real. Script prompts
- 0:24and the clot scale all in the blog post
- 0:27below. swap any photos or make it your
- 0:29first paid gig. But first, let's watch
- 0:32the full love story.
- 0:48[music]
- 0:56Couldn't
- 1:04[singing]
- 1:10those
- 1:14[music]
- 1:27>> [music]
- 1:44>> Don't run out of my foreign.
- 2:07[music]
- 2:10Hey
- 2:12See
- 2:31Hey, see what I say.
- 2:41[music]
- 2:56>> [music]
- 3:01>> Hi.
- 3:03What are you listening to?
- 3:04>> Ah, it's Orin Con.
- 3:06>> Oh, I love Oring Con.
- 3:07>> Oh, really?
- 3:12[laughter]
- 3:13>> I'm Nate.
- 3:14>> Isa. Nice to meet you. Oh, that's my
- 3:16floor.
- 3:22Holy
- 3:26Isa, wait.
- 3:27>> The first step of every AI video
- 3:29workflow is the assets, the characters,
- 3:31the locations, the props. Let's start
- 3:33from scene one. [music] So for this,
- 3:35start in Hixold Cinema Studio and name
- 3:37the new project love story. Set up a
- 3:40folder for every scene, subfolders for
- 3:43each asset before generating a single
- 3:45thing. Trust me, [music] 2 weeks from
- 3:47now when you're sitting on 400
- 3:49generations, future you will thank you.
- 3:51So, we start with the character sheets.
- 3:53Here's what everyone tries first. And
- 3:54here's why it fails. You take the
- 3:56person's photos, feed them in as
- 3:58references, and generate a character
- 3:59sheet. Let's see what we get. So, inside
- 4:01clock, I'm going to ask for a character
- 4:03sheet for my 1900s British traveler
- 4:08character. Young guy, early 20s, uh,
- 4:11with a stubble. I'm going to say use the
- 4:14attached face references [music]
- 4:15soft even lighting no shadows.
- 4:19Now I'm going to take this detailed
- 4:20prompt and inside Cinema Studio I'll use
- 4:23Cadream [music] cuz it works best with
- 4:25the faces and textures for the clothing
- 4:28and generate a couple options.
- 4:32Yeah, this is what I'm talking about. So
- 4:34the face it feels a bit off. It looks
- 4:37like a lookalike actor, not the [music]
- 4:40actual guy. the bride would totally
- 4:42notice because she always does. Now, the
- 4:45clothing texture looks realistic, but
- 4:47again, the face is not quite what I
- 4:49want. [music] You might see a weird
- 4:51texture on his face, and we don't want
- 4:53that on our main character. [music]
- 4:55So, here's what I'm going to do instead.
- 4:57It's a pretty simple hack, and honestly,
- 4:59it sounds way too dumb to work. [music]
- 5:01I built a hybrid. I kept the costume and
- 5:03the body and the overall silhouette
- 5:05using the Cadream sheet because [music]
- 5:07that part was already perfect. And then
- 5:09I just erased the head on the main
- 5:11panel. [music] And then in the portrait
- 5:12panel, we pasted the actual photo right
- 5:15in. This gets you natural looking faces
- 5:17and consistent [music] outfits every
- 5:18single time. Now for her sheet, same
- 5:20pipeline, second run. Everything runs
- 5:22exactly the same. A quick request to
- 5:24claude expanded into a full detailed
- 5:26prompt generated in Hixold and then her
- 5:29real photo swapped onto the head on the
- 5:31front panel. These exact steps are what
- 5:33keep you from [music] getting that
- 5:34generic sloppy AI look. Same rules, same
- 5:36order. The only thing that [music]
- 5:38changes is the request. All right, now
- 5:40the locations. About 70% of your final
- 5:43video quality comes from the [music]
- 5:44location. That's what makes or breaks
- 5:46your video. For our first scene, we need
- 5:48the 1900's hardware. And for that, we
- 5:50need to fit four key moments. A packed
- 5:53crowd, a shoulder bump, a long [music]
- 5:55sprint, and a leap onto a moving ship.
- 5:57So, this is what I asked from Claude.
- 5:59And this is a detailed prompt we have.
- 6:01[music] Now, I'm running a few batches
- 6:02in Soul Cinema. Hands down the best
- 6:04bottle right now for location generation
- 6:06because of the variety it [music] gives
- 6:08you. Since eight batches only cost one
- 6:10credit, I had plenty of options to pick
- 6:12from. With a scene this packed, the
- 6:14first take never survives anyway, so I
- 6:16generated several. Let's see what we've
- 6:18got.
- 6:20For the first one, gorgeous [music]
- 6:22ship, right? But look at the pier.
- 6:25There's no room for a crowd and
- 6:27definitely no room for a run. For the
- 6:30second one, no bridge [music] and no
- 6:32room between the cranes and way too much
- 6:34clutter. The extra birds and junk on the
- 6:37left will just turn into mush during
- 6:39[music] generation.
- 6:41The third one, good layout, but wrong
- 6:44lighting. That heavy yellow tint would
- 6:46leak onto every single scene. [music]
- 6:48Not our mood. Now, this this is the one.
- 6:52The long pier gives you plenty of room
- 6:53for a crowd, a crash, and a full sprint.
- 6:56[music] She can get on board, and he has
- 6:58a place to run. Best of all, the sides
- 7:00hold all the mess, leaving [music] the
- 7:02middle clean for the action. Okay,
- 7:04location locked. Let's move on to step
- 7:06three, the scenes. First and foremost, I
- 7:09don't write video prompts by hand. I use
- 7:10a skill I build and clock. It's packed
- 7:12[music] with director tricks, breaking
- 7:14down shots, timing, physics, camera
- 7:17angles. Grab the link down below. And
- 7:20all I do is describe the scene like I'm
- 7:21talking to [music] a friend and attach
- 7:23the assets. So, for scene one, the story
- 7:25is 1905, a departure to appear, Golden
- 7:28Haze. They don't know each other yet,
- 7:30and she's late for a ship. He's the only
- 7:32man walking against [music] the crowd,
- 7:34and they clip shoulders. So, I'm going
- 7:36to go into Claude and explain that same
- 7:39thing in detail. [music]
- 7:42And now I'm taking that detailed prompt,
- 7:43dropping it into Cinema Studio. For the
- 7:46model, I'm going to go with Cense 2.5,
- 7:48[music] 30 seconds, 1080p.
- 7:51And let's see the first batches. Okay,
- 7:54here's our first problem. The model
- 7:56totally missed the point of love at
- 7:58first sight. Look at this. It keeps
- 8:00turning a simple bump into a full-on
- 8:02hug. Chest to chest, even kissing in
- 8:04some of the scenes.
- 8:07Pump the brakes here. When frame one is
- 8:10already jumping straight to scene six,
- 8:12two quick fixes made the scene work.
- 8:14One, break down the physical mechanics.
- 8:17Instead of they bump into each other,
- 8:19write the exact motion. Shoulders hit,
- 8:21suitcases swing, momentum carries them
- 8:24half a step past, planted feet turn.
- 8:26When the collision becomes physics, the
- 8:28romance feels natural. Two, every
- 8:31background character is always moving.
- 8:32[music] To fix the NPC crowds staring
- 8:35into the space, forced every person to
- 8:37be mid-action in every frame. With all
- 8:39those fixes dialed in, [music] I took it
- 8:41back to Claude and got a fresh prompt.
- 8:43Let's run it again in Cance 2.5.
- 9:02>> [music]
- 9:08[music]
- 9:09>> Yeah, now that's a collision. Shoulders
- 9:11momentum, the double stop. And here's
- 9:15why CDS 2.5 really nails it here.
- 9:17[music] It understood every physics
- 9:19detail we wrote. So, fewer batches to
- 9:21the result. Okay, this one's locked.
- 9:24Let's move on to scene two. He's thrown
- 9:26into a gladiator arena against [music]
- 9:28the giant, gets tossed across the sand,
- 9:30grabs a sword, and takes the giant down.
- 9:32He looks up at the queen's box for a
- 9:34reaction, [music] but she just flashes a
- 9:36flirty look and walks away. For the
- 9:38assets, we have Nate, Isa, the giant
- 9:40opponent, and the arena background all
- 9:42organized right inside the canvas. And
- 9:44spoiler alert, I'm about to show you how
- 9:46to beat the biggest bottleneck in a
- 9:48video right now. Sizing. Let's go. So,
- 9:51the real characters will cast from the
- 9:53photos. But the giant, we need to invent
- 9:55him. And honestly, generating a madeup
- 9:57character like this is the best way to
- 9:59go. It's fast, cheap on resources, and
- 10:01since the model gets to be totally
- 10:03creative, you instantly get a ton of
- 10:05incredible options. So, watch how this
- 10:06works. I'm going to type a short request
- 10:08to clot, which is [music] going to turn
- 10:10it into a full detailed prompt. Now I'm
- 10:13going to take that prompt
- 10:15straight into Soul Cinema to generate
- 10:17the character sheet. Now let's watch
- 10:19what Soul Cinema comes back with. Okay,
- 10:21look at this one. Great buddy, but the
- 10:24chains look like melted rubber. In this
- 10:27one, the axe turned into [music] some
- 10:30fantasy decoration. Way too fancy for an
- 10:33executioner.
- 10:35And this one. Yeah, I think this is our
- 10:37guy. The body is huge and soft, exactly
- 10:40how we described it. The hood with a
- 10:43rough [music] stitching really works
- 10:45here. The chains read as real heavy
- 10:48iron, [music] heavy links, actual
- 10:50weight. Okay, locked. Let's save it into
- 10:53elements [music] as giant. And now I'm
- 10:55going to describe the scene in detail.
- 10:57Drop in all the assets. And once Claude
- 10:59hands back the prompt, I go straight
- 11:00[music] into Cinema Studio and run a few
- 11:02batches. Okay, here's a problem that
- 11:05almost broke the whole scene. Size
- 11:07inconsistency. We rode three times
- 11:09taller than [music] the hero and the
- 11:11model just wouldn't hold it. One day
- 11:13he's a monster, next day he's barely a
- 11:16head taller. Every batch a different
- 11:17giant. So, here's [music] how we fix it.
- 11:20I use Claw to convert our character
- 11:22sheet into a structured prompt, keeping
- 11:25the boy's head aligned at the giant's
- 11:27mid thigh, and then render that height
- 11:29comparison sketch in GPT image. Oh yeah,
- 11:32check this out. A basic pencil sketch,
- 11:34[music] just two outlines showing the
- 11:35hero and the giant. Now I'm going to
- 11:38feed it back as a visual reference and
- 11:40describe the fight to Claude again and
- 11:43get a [music] detailed prompt.
- 11:45I'm going to take that prompt into
- 11:47Cinema Studio. Let's run it and [music]
- 11:49see what we get.
- 11:57[panting]
- 12:06>> [groaning]
- 12:12>> Oh yeah, this is where Cance 2.5 really
- 12:15shows [music] off. Two things I noticed
- 12:17right away. First, it just understands
- 12:19the prompt better. Fewer batches to get
- 12:22what I asked for on 2.0. This fight
- 12:24would have cost me twice the
- 12:25generations. And second, the physics.
- 12:28[music] Watch his body in the fast
- 12:29movements. The dodges, the tuck and roll
- 12:31after the throw, [music] the sharp cuts
- 12:33sideways, the limbs don't actually
- 12:35morph. Now, the key decision of this
- 12:37scene, I split it in two. Part one is
- 12:39the fight, pure physics, the grab, the
- 12:42throw, the takedown, and part two is the
- 12:44acting, her worrying, her exit, his hope
- 12:48draining. Why split it? Because one
- 12:50prompt can't focus on both. The fight
- 12:52gets softer, [music] the faces get
- 12:54flatter. two prompts, full focus on
- 12:57each, then stitch them in the edit. So,
- 12:59here's my request for part two [music]
- 13:02and describe the fight to Claude again
- 13:05and get a detailed prompt. Let's run it
- 13:08and see what we get.
- 13:12[groaning]
- 13:25>> [cheering]
- 13:29>> again.
- 13:31>> And here's the finished scene. [music]
- 13:33He hits the wall, falls, and that fall
- 13:35takes us straight into the next scene.
- 13:37Now, they're both robbing a train that's
- 13:39racing toward a sea [music] bridge. The
- 13:41tunnel on the other side. We've already
- 13:43locked the assets for the scene. The
- 13:44workload is the same. [music] Let's
- 13:46focus on the videos. the biggest scene
- 13:48in the film. 16 shots, a western movie,
- 13:51a train robbery on horseback, a
- 13:53shootout, and inside all that chaos,
- 13:55[music] he's still chasing her. I gave
- 13:57Claude the beats plus their characters.
- 13:59Him clumsy and unsure, her precise, no
- 14:03wasted movements, got the prompt, and
- 14:05ran the batches. Let's watch them.
- 14:08Yeah, you can see the slop right away.
- 14:10So, in this one, he's galloping right on
- 14:13the rails. And here the jump shot is
- 14:16just not something I asked for. The
- 14:18scene is complex. A lot of shots, a lot
- 14:20of tricky actions, and the model keeps
- 14:23losing the logic of what happens where.
- 14:25Then it hit me. We've actually [music]
- 14:27solved this before. Remember the giants
- 14:29height chart? If a problem is about
- 14:31spatial logic, stop describing it and
- 14:33just draw it. So I sketch the shots by
- 14:36hand. So the first one is the slow-mo
- 14:38jump. The horse stays behind the car off
- 14:40to the side and flies in on a diagonal.
- 14:43So, here's the process. I'm [music]
- 14:45going to save each sketch as an element,
- 14:47attach them to the chat, and tell Claude
- 14:50that these are the storyboards for the
- 14:53exact shots. Claude reads the geometry
- 14:55from the drawings, the story from text,
- 14:58and nails the perfect prompt. Now, let's
- 15:01see them side by side. Look at that one
- 15:04ugly sketch. And the special layout
- 15:06finally started to work. [music] The
- 15:08horse stays off the rails in every take.
- 15:10The diagonal jump lands exactly like the
- 15:13sketch. The geometry held from these
- 15:15batches. We pulled the best pieces into
- 15:17the [music] edit. And the same lesson as
- 15:19the giant. If it's about space, draw it.
- 15:21Okay, here's the full scene.
- 15:53New era, new tactic. We have pirates.
- 15:56He's a captain now, and he's in love
- 15:58with the captain of the enemy ship.
- 16:00[music] The only way to get close to her
- 16:02is to attack her ship. This scene needed
- 16:04more assets than any other. Two
- 16:06captains, a dualist, the deck, a spy
- 16:08glass, [music] and the whole doors, plus
- 16:11one special asset that solved a unique
- 16:13problem. One more thing about the
- 16:14character sheets before we shoot. I run
- 16:16the same sheet prompt through [music]
- 16:18three different models, Cream 5 Pro, GBT
- 16:20Image 2, and Nana Banana, and compared.
- 16:23So, for him, Cream [music] 5 Pro, the
- 16:26costume has the most texture and wear,
- 16:28and it stays consistent across all three
- 16:30panels. For her, GBT Image 2 has the
- 16:34better curls. Full, natural, identical
- 16:36from every single angle. The other
- 16:38models couldn't hold them. So, here's
- 16:40the takeaway. Every model brings its own
- 16:43strength. [music] Choose the best one
- 16:44for the job. Then, the next problem
- 16:46showed up. A pirate battle needs [music]
- 16:48pirates, and my background characters
- 16:50kept generating as random mush.
- 16:52Different faces every shot, sometimes
- 16:54barely human. So, I cast them as extras
- 16:58like a real movie would. I lined up 10
- 17:00different pirates against the gray
- 17:01backdrop and packed them into a single
- 17:03element, the crew. Now, every deck shot
- 17:06pulls from the exact same characters,
- 17:08assets locked, and now I'm handing
- 17:10clawed part one of our scene. And this
- 17:12is the prompt that I got. Let's run a
- 17:15couple batches and see what we get.
- 17:16[music] I mean, I love how this turned
- 17:18out. I just grabbed the best shots from
- 17:20each render to build the scene. One
- 17:22prompt gave me four takes and [music]
- 17:24three made the cut. I'm going to be
- 17:26honest with you. A perfect 30 secondond
- 17:27AI render just doesn't [music] exist.
- 17:30You generate raw footage, then stitch
- 17:32the scenes together in the edit. This
- 17:34scene is split in two because the full
- 17:36sequence doesn't fit into [music] one
- 17:37generation. Part two is short, only 15
- 17:39seconds, and is my favorite beat in the
- 17:41whole film, the duel that never happens.
- 17:44Here's the prompt. Let's see the
- 17:45results. [music] Again, four batches
- 17:47from one prompt. I took the battle
- 17:49behind him from this generation and the
- 17:51whole duel from this one. The action
- 17:54just better here.
- 18:07[screaming]
- 18:25All right, on to the next scene. New
- 18:27setting. Old Italy night, fireworks,
- 18:30[music] a mass ball across the palace
- 18:32square. Both in burgundy and both in
- 18:34masks. This is the closest they've ever
- 18:36been, and neither one knows who the
- 18:37other is. At this point, I genuinely
- 18:40feel like I'm directing a romance drama.
- 18:42[music] Starting with the assets, full
- 18:44carnival gear, masks, and the crowd
- 18:47[music] square, all linked in the post
- 18:48below. And notice the location. This
- 18:51time I generated it already alive with
- 18:54the people. I did this because a
- 18:56location [music] without the crowd here
- 18:57kept breaking. So I generated a quick
- 18:59test to see if the crowd would hold. And
- 19:02this is what it looks like. There we
- 19:04[music] go. Beautiful. The square is
- 19:05alive. The festival feels real. And the
- 19:08prompt stayed clean and simple. This is
- 19:11the whole trick. So assets locked. Now
- 19:14my next prompt. Let's generate. Okay.
- 19:17Okay, I tested it as a one 30 secondond
- 19:19prompt and honestly it held but this is
- 19:21the pref finale. The first time they
- 19:23touch acting needs room and inside one
- 19:26giant prompt it just doesn't work. So
- 19:28one more message to claude. Let's split
- 19:31the scene above into three prompts 15
- 19:34seconds each. Prompt one. From the
- 19:37beginning up to the moment he breaks out
- 19:40of the circle [music] of dancing girls.
- 19:43Prompt two. The hero runs through the
- 19:46crowd,
- 19:48sees the girl across the road.
- 19:51The carriage passes, the hero [music]
- 19:53disappears.
- 19:55She searches for him, turns around, and
- 19:58walks into the thick of the crowd behind
- 20:01her. And prompt three, the girl walks
- 20:04through people in the foreground looking
- 20:07around.
- 20:09Now, I'm going to generate each prompt
- 20:11separately. [music] And this is why you
- 20:14split the takes. Just look at their
- 20:15faces. You can actually see him getting
- 20:18lost in the dancers and you really feel
- 20:21her scanning the crowd in one long prom.
- 20:23These beats felt completely rushed.
- 20:25Breaking it up gives it room to breathe
- 20:27and it's [music] a total game changer.
- 20:39>> [cheering]
- 20:55>> Okay, that was the carnival. Now the
- 20:58finale. This is present day, the campus
- 21:00elevator because that's the real story.
- 21:02The real Nate and Isla met in college.
- 21:05[music] and everything you've seen so
- 21:06far, that's what was playing in his head
- 21:08every time he tried to work up the nerve
- 21:10to talk to her. The assets, him as a
- 21:13student with a backpack on watcher, her
- 21:16in a beige cardigan, the elevator,
- 21:18[music] and headphones, all again in the
- 21:20blog post. But there's one more key
- 21:22asset [music] that you won't ever see on
- 21:24screen. This whole scene lives or dies
- 21:27on sound. And here's how we pulled it
- 21:29off. My friend Nate recorded a quick
- 21:31voice [music] memo humming our main
- 21:32theme. Just that simple. D melody.
- 21:36Without it, the model just hallucinates
- 21:38a new melody every take. But with it,
- 21:41the character hits those [music] exact
- 21:42notes on point. And the best part, works
- 21:45with any track. Just drop your favorite
- 21:47[music] song in as a reference and copy
- 21:49the trick. Okay, first let's watch what
- 21:51happens if we don't attach the audio
- 21:53reference.
- 21:54>> Oh, really?
- 21:59[laughter]
- 22:01>> My brother's just terrible at singing.
- 22:03off key zero rhythm. Now let's try it
- 22:07with the audio. So in claude I'm going
- 22:10to drop in my request.
- 22:14Take that new prompt into C dance and
- 22:17upload the audio into the prompt here
- 22:20and let's run a couple batches.
- 22:24>> Hi. What are you listening to?
- 22:26>> Ah, it's Oring Conan.
- 22:28>> Oh, I love Orcon.
- 22:29>> Oh, really?
- 22:34>> [laughter]
- 22:35>> I'm Nate.
- 22:36>> Isla. Nice to meet you. Oh, that's my
- 22:38floor.
- 22:39>> Boom. First take, perfect execution.
- 22:41[music]
- 22:42So, audio references aren't a gimmick.
- 22:44They're how you force the model to get
- 22:46the [music] acting right. So, that's the
- 22:48whole pipeline. The skill, the script,
- 22:50the prompts, everything I used are
- 22:52linked below. And better yet, we've also
- 22:54[music] created a custom app where all
- 22:56you have to do is upload your photo and
- 22:58the bride's photo, and you get the full
- 23:01exact same love story built for [music]
- 23:03you. So, go try it out. And as always,
- 23:06if you find this video helpful, hit the
- 23:08like button, subscribe, and I'll see you
- 23:10guys in the [music] next one.
About this transcript
This page contains the full transcript of How to Make a Wedding Love Story Using AI (Full Tutorial) by Higgsfield AI, generated from the public captions YouTube serves with the video. The transcript has 3,310 words across 511 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.