**3-Step Workflow To Master Ultra Realistic AI Commercials (Full Step-by-Step Workflow) — Transcript & Summary | SozAI**
Source: https://sozai.app/transcript/ai-commercial-workflow-step-by-step/

Higsfield AI tutorial showing a step-by-step workflow for creating a 30-second AI delivery commercial: character sheets, locations, props, and directing with Cinema Studio and Soul models.

## Key Takeaways

- A consistent AI commercial depends on locked character references, location layouts, and prop assets built before video generation
- Editing models help refine details while preserving lighting and composition
- Planning seating, camera paths, and transitions in advance reduces errors in the final video
- Iterative generation and asset organization are essential when producing AI commercials

## What the video covers

- Introduces a 30-second AI-generated delivery commercial featuring three friends across three countries ordering from one app
- Sets up a project structure in Higgsfield's Cinema Studio with organized asset folders
- Creates character sheets for Micah, Jenny, and Alex using Claude-generated prompts and Soul Cinema, with a plain gray background trick for consistency
- Uses Cream 5.0 Pro for image editing to refine skin texture, lighting, and remove unwanted details
- Builds locations (apartment, Mexico, Italy, Japan) using image references and layout edits to keep camera space clear
- Adds props such as phones, delivery bags with logos, and grouped extras like musicians
- Directs the scene with attention to seating, reactions, camera movement, and transitions between locations
- Uses color-coded seating and anchor elements to keep reverse angles consistent
- Applies sound design tips, such as bringing in ambient audio before visual reveals
- Discusses total cost including failed generations and the time investment (two weeks, around 400 generations)

## Chapters

1. 00:00 Intro: The Food Delivery Ad Concept
2. 01:09 Character Design: Micah, Jenny, and Alex
3. 03:03 Building the Apartment Location
4. 05:10 Mexico, Italy, and Japan Locations
5. 07:00 Props, Phones, and Extras
6. 08:00 Directing the Opening Scene
7. 10:00 Seating, Reverse Angles, and Transitions
8. 13:00 Pack Shot, Sound Design, and Wrap-up

Answers

## Questions about this video

How does the creator keep the characters consistent across scenes?

They generate a character sheet with a close-up portrait and front and back outfit views on a plain gray background, then save the best face as a reference. That reference is attached whenever new scenes are generated, so the character's look stays consistent.

Why are image references used for locations instead of text prompts alone?

Image references lock down lighting, layout, and background before video generation, similar to a live-action shoot. The creator also uses 3/4 angles because they give camera movement more depth to work with.

What tools are used to refine generated images?

Claude is used to write detailed and edit prompts, Soul Cinema generates the initial images, and Cream 5.0 Pro handles image editing for details like removing freckles, softening lighting, and clearing layout clutter.

## Full Transcript — Download SRT & Markdown

00:05

Speaker A

Okay, we need food. Let's check Higos. Ooh, pasta. Wait, look at these tacos. Tell me you don't want that.

00:19

Speaker A

Actually, I've been craving sushi. Why choose? Let's get all three. Higos, from anywhere to your door.

00:30

Speaker A

I directed this 30 secondond commercial in Cance 2.5. Three friends, three countries, and one delivery app. I'll show you how I built the characters, planned the transitions, and fixed the shots that went wrong. And at the end, I'll show you what it cost, including

00:44

Speaker A

the failed takes. First, here in Kicksfield, go to Cinema Studio, create a new project, name it delivery ad. On the left, there's a folder bar. I'll create a folder and add subfolders for each asset. Trust me, 2 weeks and 400

00:58

Speaker A

generations later, you'll be glad you did this. Okay, let's start with the characters. For this, I'm using Soul Cinema. It's my top choice right now for characters, and it's affordable enough to try a bunch of options. But first,

01:09

Speaker A

I'm going to throw in the basic concept into Claude to get a detailed character sheet prompt. I'm going to start with Micah. She's the one who suggests checking the app. So, I want her to feel friendly and playful. Ginger hair and a

01:21

Speaker A

messy bun, small glasses, just a casual look. I'm asking for a character sheet with a close-up portrait, plus her outfit from the front and the back. By the way, here's the huge hack. I'm using a plain gray background so her face and

01:33

Speaker A

clothes are easier to see. No visual noise, total focus on the character, and instantly more usable results. Next, I'm going to drop in that prompt into Soul Cinema and let's take a look. So, all of these have the same issue. The skin

01:45

Speaker A

looks too smooth and also the lighting feels a little too harsh for me. Let's choose this one. But I want to remove the freckles on the face. So, for this I'm going to use Cream 5 Pro, the best

01:56

Speaker A

model right now for image editing. I'm going to ask Claude to edit the character sheet or give me a detailed prompt. Yeah, that's better. So, we got softer lighting and you can actually see some skin texture. She has more of a

02:09

Speaker A

commercial look now. Okay. And one more thing with those neck down crops, Cedense only has one face to use. The other panels are just for clothes and proportions. This is a must if you want consistent characters. Now, I'm going to

02:20

Speaker A

save her as character Micah. This is the reference I'll attach whenever we generate the scenes. Next, we have Jenny. As a director, I want her to feel confident about what she wants. Long black hair, round glasses, and a cream

02:33

Speaker A

hoodie with a beanie underneath. This was my request for her. All right. And throwing that detailed prompt back into Soul Cinema. Since I'm aiming for something super unique, I want to pump out as many options as possible. And

02:44

Speaker A

honestly, this is where Soul Cinema really shines. Eight images cost only one credit, so it's the absolute best tool when you're exploring different character concepts on the cheap. And here are the options we have. H, this one's missing the beanie. I want both

02:58

Speaker A

layers. And there we go. Beanie, the hood. I got the calm smile. This will be Jenny. And the last one is Alex. He gets dark curls, an easy smile, and a red checkered shirt. Let's see what we get.

03:10

Speaker A

Not going to lie, these all look kind of solid, but I think this one right here is my favorite. Now, now that our cast is locked, time to move on to building the apartment. For the apartment, I want

03:21

Speaker A

somewhere that these three would actually hang out. Warm lamps, blue light throughout the windows, records lying around. And here's the trick. I'm using image references for this setup.

03:31

Speaker A

Why bother? Simple. Just like a real live action shoot, you want to lock down your lighting, layout, and background before hitting generate. For this scene, we're mixing warm, cozy indoor light with that cool blue hour coming through the window using an open kitchen

03:46

Speaker A

background to give the shot tons of depth. I'm building the layout around the action. I'm going to drop all these references into Claude and ask for location. And a pro tip, always go for 3/4 angle. That added depth gives camera

03:59

Speaker A

movement something solid to grab onto and gets you way better renders than a flat head-on shot when making the videos. Now, heading back to Soul Cinema for our location pass. And seriously, this model just gets it. The colors, the

04:11

Speaker A

mood, and that rich cinematic feel are always on point. All right, let's see what we get. Out of all the generations, I think this angle is definitely the winner. But here's an issue. This foreground cabinet is taking up way too

04:24

Speaker A

much space. That can be a problem later when we make the videos. And the coffee table also needs a little cleanup. I want to leave room for the bags when the delivery arrives. So, I'm going to ask Claude for an edit prompt, fix that

04:37

Speaker A

layout, shrink the cabinet, and clear the table. Keep the lighting and the rest of the room the same. Now, I'm going to take that location along with a new prompt straight into Cedream 5.0 Pro and make a new batch. Oh, yeah. That

04:50

Speaker A

opens up the room right up. That's why I use Cream here. The model is good at editing certain details while preserving the lighting and texture of the original. Now, we have the chair and a clear view of the table. I'm going to

05:02

Speaker A

save this location as location flat 2. And next up, we have Mexico. I want hard midday sun, market stalls, and a takaria overlooking a valley. We need a table for the three friends with enough space for the musicians nearby. I'm going to

05:17

Speaker A

give that brief to Claude, then take the prompt. It writes into soul cinema and generate four options. I think I like this one. The market fills one side of the shot and the valley opens up behind the building. I want to keep that view.

05:30

Speaker A

But the table needs some work. We need three clear seats with a plate in front of each person and nothing blocking the camera. So, I'm going to ask Claude for another edit prompt, then use it with a selected image in Cream 5 Pro. Yeah. So,

05:44

Speaker A

this is the layout I wanted. The side close to the camera is open. So, we can see all three places at the table. Each chair has a plate in front of it. And the smaller salsa bowl leaves more room

05:55

Speaker A

for the tacos. And by the way, I added the food directly to the location references. Each friend gets a seat with a plate in front of it, so I've already planned where they'll sit, and I don't need to generate separate food

06:06

Speaker A

references. For Italy, I'm asking Claude for an edit prompt. Three chairs, a plate of pasta in front of each one in the same light and composition. And here's the result. The food is already part of the location, and the table is

06:18

Speaker A

set for all three of our friends, right by the canal with the gondelier. So, you instantly know where we are. Japan was the easiest to make. Now, let's move a little bit faster. All of the locations are saved on the canvas. The exact

06:32

Speaker A

prompts and location images are linked below so you can try them yourself. All right. Now, let's generate the phone, the delivery bags, and the extras. I'm going to make the props in 5.0 Pro. The phone passes between all three

06:43

Speaker A

characters so it gets its own reference. Same for the packaging. I want the Higos logo facing the camera in the final shot. And here's a shortcut for extras.

06:51

Speaker A

Build the whole group in one reference. I made all four musicians together in one image. And here's the catch. Grab all of this, the prompts and the quad requests in the description to try them yourself. And a quick pause before we

07:03

Speaker A

move on. I checked the numbers this week and about 80% of you watching right now are unsubscribed. Can you please do me one favor? Check that button. That's it.

07:11

Speaker A

It cost you nothing. But every subscriber means we get to go way bigger on the next build. Now to the most interesting part, directing our commercial. Everyone's starving. Micah's stomach growls. Jenny sees a chicken leg instead of a flower and Alex starts

07:25

Speaker A

behind the cat. Micah opens Higos. One tap takes them to Italy. So, Alex gets the bean bag beside Micah sofa so the camera can move between them when she interrupts. Jenny gets the armchair and the coffee table needs room for the

07:40

Speaker A

delivery bags. The transition matters here. That tab is our cut to the next country with the camera still moving.

07:46

Speaker A

Let's build that into the prompt. First, I use my special prompting skill I build in Claude. To load it, click Customize, then go to skills and hit the plus button. Then, click upload a skill. And that's it. Now, I'm going to activate

07:58

Speaker A

the skill and describe my scene. The skill organizes it into shots with dialogue timing and camera moves. I'm also going to attach the reference images and label them with the same names I'll use in the prompt. Italy is

08:11

Speaker A

here because this scene needs to end in Italy. So, this is where directing gets specific because I have to think about where people sit, when they react, and what the camera does. This covers almost 17 seconds. I've timed Micah's

08:24

Speaker A

interruption in the tab so they land where I want them. For Jany's gag, I want a clear view of her face and the flower. The camera moves past the table and shift focus to the flower as it changes. Micah cuts off Alex's cat joke

08:37

Speaker A

halfway and the focus snaps to her instantly. That is what makes the joke hit. And here's the transition. Cut on the tab. Keep micas framing and the camera movement the same. Only the location changes. That's the match guide. For video, I'll be using CD 2.5.

08:51

Speaker A

It's hands down the best model right now. Here inside Cinema Studio, I'm just going to paste in my prompt. And as you can see, all the references get attached automatically as elements. The steak comes to about 200 credits. Let's run

09:04

Speaker A

it. Okay, I like the flower gag and Alex with a cat is exactly the ridiculous vibe I wanted. Keeping those reactions, but look where Alex ends up. He's on the sofa next to Micah and I wanted him on

09:24

Speaker A

the bean bag. Looking back at the prompt, I gave him the cat, but never actually assigned a seat. That's on me.

09:30

Speaker A

I'll keep the action, make the position explicit, and add a short wide shot before the close out. Here's why. That extra shot of the whole apartment is the main key frame for the scene. It locks in where everyone sits and gives the

09:42

Speaker A

model clear visual anchors in the background. Getting all three friends into positions right away keeps their seats consistent when we cut to close-ups. I'm also putting Jenny back in the armchair for this version. The flower gag stays low. This time I'm

09:56

Speaker A

specifically pointing out that Alex is on the bean bag and connecting each seat to a landmark in the background.

10:02

Speaker A

Skylight behind Micah, shelves behind Jenny, the red lamp behind Alex. I know that's a lot of detail, but those landmarks give me something to check as the camera moves. Bringing that updated prompt back into sea dance with the same

10:14

Speaker A

references. Let's see how it looks now. And now he's on the bean bag with Micah's sofa right beside him, even in the closeup. There we go. That's where I wanted him. And that transition into Italy. Love it. keeping that part. Now,

10:32

Speaker A

keep in mind that you don't have to use every shot you generate. I added the white to establish the layout for the whole generation, but I will leave it out of the final edit and use the strongest moments from each take that

10:43

Speaker A

keep the seating consistent. Next, Alex gets the phone. Micah is about to eat, and that's when Alex decides he wants tacos instead. He grabs the phone just before she takes a bite of her pasta.

10:53

Speaker A

Again, I'm explaining my idea to Claude. Let's break down the transition. Before the phone grab, I want to give Italy a moment. A gondel passes. Alex gets up to look, then sits back down. The location gives him something to react to. These

11:06

Speaker A

are three beats in the same sequence. Follow the phone, push towards Alex, then cut on his tab. His pose matches across the cut and the camera keeps moving as Italy becomes Mexico. Now I bring the full prompt into seance. Let's

11:19

Speaker A

see what we get. Wait, look at these tacos. Tell me you don't want Alex keeps filming with his left hand and eats with his right. He's trying to sell his friends on the tacos with his mouth full. Now, let's look at the reverse

11:36

Speaker A

angle. I want the musicians in the foreground with Alex eating his taco at the table behind them. Here's the prompt from the first takes. It described the camera moving behind the band looking toward the table. But look at what came

11:48

Speaker A

out. We're behind the musicians looking back at the building with the hills sitting behind our friends. Even though I mentioned the hills earlier, sedans kept putting a wall behind Alex. So, I had to spell out exactly what we should

12:00

Speaker A

see from this specific angle. So, here's the direction choice. Keep the band where it is, put the camera beside the building, and point it out toward the valley. In the later prompt, that becomes much more specific. The building

12:12

Speaker A

stays beside the camera, the table is beyond the musicians, and the hills are in the distance. And there it is.

12:17

Speaker A

Musicians in the front, our friends at the table, and the hills behind them like we wanted. And a quick pro tip, anchors are very important in locations, especially when you have reverse angles.

12:27

Speaker A

Use them to tell the model how to position or move the camera around those anchors. Next up, Jenny grabs the phone, and I'll show you how to mask the cut.

12:36

Speaker A

So, the jump to Japan feels totally seamless. Jenny wants sushi, so we're off to Japan. The move into Mexico pushed in on Alex. For this shot, I want a sideways reveal to switch up the pacing. Jenny's still tapping the phone,

12:50

Speaker A

but the camera slides right behind Micah. And I'm making a DP move here, using her back to hide the cut. I'm giving the position of each character and the detailed action. Let's check it out.

13:01

Speaker A

Actually, okay. Jenny taps before Micah's back fills the frame. That's where I hide the cut. The camera keeps moving, so it feels like one continuous move into Japan. Here, I also make another director's trick. I'm going to cut the

13:15

Speaker A

music and bring in Japan's ambient sound first. That way you hear the new space before the reveal. That's directing.

13:22

Speaker A

Deciding who sits where, how the camera moves, where to hide the transition, and when the audio changes before you make the final render. Now, let's look at the next shot. For Jenny's sushi line, I'm cutting to a top shot with a smooth

13:34

Speaker A

clockwise rotation. We can see the food and all three friends while the camera keeps the scene moving. And let's hit generate.

13:41

Speaker A

I've been craving sushi. Three friends, but a fourplay setting, even though the prompt explicitly asks for three. Seems thin in an extra spot.

13:48

Speaker A

So, I'm giving the model a map of exactly what belongs where. In Cream, I specify the seating, the plates, and the phone on their own. I picked this one because the layout breakdown is so clear. Every figure gets its own color

14:01

Speaker A

code, which I then map straight to each character in the video prompt. So, in this case, orange is Jenny on the left, blue is Alex at the top, and green is Micah on the right. Three individual plates, two shared dishes, and the phone

14:15

Speaker A

face down beside Jenny. For this overhead shot, I also need room around their heads as the camera rotates. Now, you can use our blocking skill to make maps like this. It helps Claude write the map prompt and explain how to use it

14:27

Speaker A

in the video. So, the model has clear positions before we spend credits on another take. That skill is in the description, too. I'm going to drop in the map, my blocking skill, and the original references straight into Claude. The map controls placement. The

14:40

Speaker A

photos bring in the characters plus the room, and those colors are just labels. Jenny has already put the phone down.

14:46

Speaker A

She opens her hands toward the sushi and says the line. The plate stays put. Now I'm going to bring the revised prompt back into seed ends and attach the map alongside the other references. Let's see what we get.

14:57

Speaker A

I've been craving sushi. Yeah, now it's way better. This is exactly the table arrangement I wanted.

15:03

Speaker A

By the way, our Discord community just passed 200,000 members. And to celebrate, we're running a contest with a prize pool of 200,000 credits. Link in the description. All the details are in the server. Okay, we'll lock in the stop

15:15

Speaker A

shot and move straight to the delivery and the ads closing shot. Everyone's made their pick. Now we need to bring the food home. Micah's got the simple solution.

15:23

Speaker A

Why choose? Let's get all three. On the doorbell, we cut back to the apartment. Deliver is here. Everyone's eating. And yes, the cat gets some, too.

15:33

Speaker A

I want to start close on the Higgos packaging brand first. Then the camera pulls back and drops down to the cat with our friends eating in the back.

15:41

Speaker A

That's our pack shot. One move, four seconds. The first brief asked for one bag, but I ended up using two bags and a cup holder. The printed logos face straight to the camera so that the opening close-up gives a clean, clear

15:53

Speaker A

view of heo straight away. And the eating part I described in detail, lowering the head, takes a bite, chews.

16:00

Speaker A

I wanted something actually happening right as the camera move stops. I also added some voice over right from the beginning. Here's the take a cat. from anywhere to your door.

16:10

Speaker A

He goes first, then down to the cat. And after that look Alex gave him earlier, he's earned his dinner. I pulled the best moments from these takes into the finished commercial you saw at the start. For this entire ad, I spent

16:22

Speaker A

around 2100 credits, including a few rounds of iterations. And as you get better at directing these shots, you can spend less time and fewer credits on retries. For context, professional liveaction production in the US with actors and a full crew can cost tens of

16:36

Speaker A

thousands of dollars, so you're definitely saving me here. All the materials are in the description. Grab the prompts and give this workflow a try. Start with one short scene, fix what's off, and build it into an ad for

16:47

Speaker A

your portfolio. Use it to reach out to your first client, then show us what you made in Discord. And as always, if you found this video helpful, hit the like button, subscribe, and I'll see you guys in the next one.

Topics: AI commercial Higgsfield AI Cinema Studio Soul Cinema character consistency AI video workflow Cream 5.0 Pro AI filmmaking generative video food delivery ad

Study this video

- [Flashcards from this video](https://sozai.app/tools/flashcard-maker/?from=transcript&id=89903)
- [Quiz on this video](https://sozai.app/tools/quiz-generator/?from=transcript&id=89903)
- [Crossword from this video](https://sozai.app/tools/crossword-maker/?from=transcript&id=89903)
- [Word search from this video](https://sozai.app/tools/word-search-maker/?from=transcript&id=89903)
- [Study guide from this video](https://sozai.app/tools/study-guide-maker/?from=transcript&id=89903)

Free, made by AI from this transcript. No signup.


---
This is the markdown twin of https://sozai.app/transcript/ai-commercial-workflow-step-by-step/ — the same content, without the markup.
Published by SozAI (https://sozai.app). Reuse and quotation are allowed with attribution and a link back.
Machine-readable index: https://sozai.app/llms.txt · data API: https://sozai.app/api/