Scaling Fashion Lookbooks

Transcript

Scaling AI photoshoots

Thanks for having me. Today we're going to cover how to scale up lookbooks, editorials, social content, and campaigns for fashion, e-commerce, and direct-to-consumer brands. The goal is to build a methodic system that will allow us to repeatably and consistently simulate an on-brand photoshoot day: to bring up all the creative knowledge, brand requirement and implement the brief designs into an automated system.

Outline — what we'll cover today

Let me quickly outline what we'll cover today. First, we will review the current state of traditional photoshoots, and how things look today. Then we will see what AI photoshoots look like today, and see few examples for major brands who have implemented AI. Then we will dive into brand identity and how we can translate it into AI: how to apply it, how to work as a creative director inside Astria, and how to use real-world references. From there we'll get into prompt editing and refinement — making sure you consistently get strong outputs with fewer bugs and artifacts. Once stable and ready, we will templatize our work so that you can run looks faster — that's the core of the "scale up." We'll also cover the Describe tool for bringing existing client assets into the AI world. Finally we will deliver a digital product to a client: not just images, but a template the brand's studio can run themselves.

Traditional photoshoot

Let's start by reviewing how do these currently work pre-AI? What does does traditional photoshoot look like? Usually this would involve logistics such as coordinating models, locations, shipping inventory, making sure product arrives from production to the studio on time. Stylists, makeup, hair — all of this logistics exists, ultimately, to serve the brand's DNA. The entire photoshoot is there to convey the brand's identity, and we need to somehow implement and communicate that. Now let's see how we can do that with AI.

Who this is for — brands & verticals

Before we go further, let me clarify who this is actually for. On the brand side, we work with three types of business. E-commerce — companies selling product online. DTC brands — going directly to the consumer, with or without a store. And traditional retailers who need scalable visual content for their catalogs. The verticals where this really shines are fashion and apparel, footwear, outdoor and professional gear, jewelry, glasses, and beauty. If your product lives on a model or on a body, AI photoshoots are going to change how you produce imagery.

Real brands already using AI

Let's see a few examples of brands already using AI. SACKS rebuilt their entire homepage. The hero imagery, the model shots, the product page lookbooks — all done by the professional photographer Shai Yechezkel & Miki Turgeman who have formed an AI studio called CI898. Shai and Miki helped steer the brand's creative direction together, built the brand's avatars and template set, and then ran hundreds of looks through it. They're already done two collections with AI. Ronny Kobo is another example of a Paris based designer. You can see the difference in styling and how the branding gets applied translated into the visuals. Finally the big players such as Mango, H&M, Zara — are all doing the same thing, producing both campaigns and lookbooks with AI. The world is already implementing AI. What remains is to direct it well, scale it up, and do it precisely and custom to each brand's identity.

Brand identity in Astria — three real implementations

Here's how we actually implement brand identity in Astria. Every brand has its own visual language — its own hex colors, its own models, expressions, and atmosphere. Off-the-shelf templates don't work because a brand will always say "this is beautiful, but it's not me." So we build a workspace for each brand, with their models, backgrounds, lighting, and template set defined once — and reused across every collection. Here is an example of fashion brand modeled after Zara — editorial fashion, outdoor shoots, European aesthetic. A set of templates across casual, formal, and nightwear. Great Sports — Adidas-style. Higher intensity expressions, athletic energy, a completely different set of models and templates. And America Basics — a Uniqlo-style brand. Soft neutrals, friendly expressions. Clean, everyday. Completely different DNA from the other two. For each brand we deliver a workspace with a set of templates — one per garment type. The studio uploads a product and gets the full lookbook in a single click.

The platform is the partner

Here's what we've learned working with brands: they need constant help keeping up with AI. Brands are busy with their core business: design, logistics, fashion. They do not have time to write prompts, and they don't need to. That's exactly what we're trying to solve. Six months ago we used Flux LoRA. Then it was Nano Banana. Today it's Nano Banana 2. In two months it will be Nano Banana 3, Seedance or WAN 2.7. The brand doesn't need to know. The brand receives a template. The brand's studio can then upload a simple packshot photo, and get a full lookbook in a single click. The framework was defined for them: the models, the lighting, the framing, and so now they can scale it up and run looks in bulks. Sometimes the brand may way to do small customizations. The Brief box allow to specify a customization like "this time she's sitting on an armchair," or "make her pose more dynamic" — and it runs the entire template which keeps results consistent and tell a coherent story. Finally, it's important to explain that brands need a set of templates — one for jackets, one for shoes, one for outdoor shots. Each garment type needs different crops, framing and poses. The brand can then pick a template, and upload a bulk of garments to run the lookbook in a single click.

Your turn — start creating

Alright — your turn. First thing, let's open a Workspace. Think of it as a shared room — your team, creatives, CMOs, photographers, all working side by side. Every prompt, reference, and template you create shows up live for everyone in there. To set one up, click the Workspace selector at the top-left, next to the logo, and pick "New workspace". Name it after the brand you're working with, and hit "Create". Now — pick a brand you actually love. Skip the made-up exercise and grab something real. A designer you admire, a brand you've worked with, maybe something local you've been watching. Pull up their current material. Look at the poses, the faces, the hair, the makeup, the expressions — what makes the look theirs. Let's get our hands dirty.

Open a Workspace for the brand

Let's start by opening a new workspace. Click the Workspace selector next to the logo, pick "New workspace", and name it after the brand you're working with. I'll name mine Bobo Choses.

Brand research — what makes the brand recognizable

Here's the Bobo Choses website. Let's think about the brand's identity. How does the lookbook look — white studio, gray, concrete floor, or outdoor? How do the models look — age group, expression, hair? And the poses — strict front, back, 45-side, detail view, or something more dynamic? The adult faces have a neutral expression while the kids look a little brighter. The women's hair is parted, the skin is pale. Adult ages are around 22 to 26, kids around 8. Now let's examine the lookbook itself. On the kids' product page, a medium upper-torso waist-up view with very little headroom. The full-body view leaves a lot of whitespace, with socks and flip-flops, 45-degree front, hands relaxed at the side. The women's lookbook follows the same pattern — medium shot, upper torso, head cropped, waist up. The full-body view has plenty of whitespace, 45 degrees, legs in a V, hands at the side. And a dynamic shot rotated, with hands behind the head. The aspect ratio across the lookbook is 3:4 — quite common. Now let's go back to Astria and simulate a photoshoot day. We'll try a kids lookbook — one of the main pain points for brands, and where they're usually happy to start migrating to AI. For starters, let's get a baseline image to get things going.

Technicalities — model, resolution, aspect, count

Now for a few technicalities. Model — make sure Nano Banana 2 is selected. Resolution — 2K for iterating quickly. Once you're happy with the look, switch to 4K. Aspect ratio — 3:4. Most lookbooks are 3:4. And for quicker iteration, two images per prompt. Later, bump it to four if you want more variation.

Build the look — Lookbook references one cube at a time

Let's click the Lookbook button and start adding references. We'll go through each cube — one at a time. Background. The Background cube has a color picker built in. I'll pick white — clean studio look, good for product images. Pose. Click the Pose cube and the picker opens. You'll see options like back view, front pose, full body, 45 angle, back medium, front medium. I'll go with Front medium — straightforward, waist up, classic catalog framing. Face. Click the Face cube. Same picker pattern, but here we get the public library — every face was generated by Astria and is free to use. I'll search for "boy" and watch how the list filters down. Felix looks right for our look. Top. For garments, the fastest path is to copy a packshot straight from the brand's site. Right-click copy, then back to Astria, focus the prompt box, and Command-V. The image uploads instantly and lands in the Top cube. Bottom. Same flow with the shorts. Footwear. The public library has shoes too. I'll search for "flip" — and there's our Minimalist black flip-flops. Click, and it's added.

Refine prompts, describe, templatize

Most bugs you'll hit aren't the model being wrong — they're conflicts between the crop you asked for and the references you loaded. A medium shot with a shoe reference tries to cram a face and shoes into one frame and distorts. Keep the crop, the references, and the prompt telling the same story. When you need finer control than the Lookbook gives you, go to the prompt box. A handful of high-impact keywords goes a long way — diffused lighting, realistic skin, soft expression. That's where brand DNA lives: serious or pleased, high-end or low-cost. The Describe tool closes the loop between references and prompts. Let me grab a reference from Unsplash — a search for fashion photography pulls scenes close to what we want. Copy one, back to Astria, click Paste from clipboard, then Describe. The tool extracts a prompt, but it listens to the references you already have and avoids describing them. No more white shirt in the prompt fighting a yellow dress in the reference. Once a look works — model, pose, lighting, expression — stop iterating and templatize. Turn the pack into a template. From there it's a digital product: the brand's studio picks a template, uploads a packshot, and runs the lookbook. Build once, scale forever.

Deliver the template — the brand runs at scale

Once a look is working — the model, the background, the lighting, the framing — we stop iterating and lock it in. Open the hamburger on a generation, click Template, and save a New Template. From that point, no matter what garment you drop in, it will keep the model, the background, and the lighting you originally defined. This template is what we deliver to the brand. It's not a folder of images; it's a recipe. The brand opens My Templates, picks the one we built for them, swaps in a new garment — and everything else stays consistent: the same angle, the same model, the same scene. The template itself holds a collection of prompts — front, side, back, detail. That's what makes this scale. A brand gets a hundred shirts from the manufacturer; they select thirty at once, click Generate, and the template runs them across four models in the same angles and the same look. That's the end of the workflow: we build the template, the brand runs the lookbook.

Wrap-up

The short version: train the brand, ship the template, let the brand run the looks. Creatives stay in the directing chair — the template does the cranking. If that model fits what you're trying to build, start at astria.ai. Thanks for listening.