The Studio: AI models and photos

How to create a model, generate product photos with it or a main photo, make videos — five seconds or up to 30, with a script written by AI — review the proposals, and what label the images Bruxel generates carry.

What the Studio is

The Studio generates photos of your products from your real photos. It always starts from what you already have: it doesn't invent a product, it stages it.

• Models: fictional people you create once from a casting and reuse in as many photos as you want.
• Photos with a model: your product in that model's hands, based on one to three real photos of the product.
• Main photo: a studio version, with no people, derived from a real photo. It's the only one that can become the product's main image.

Creating a model

In Generate → Models, an administrator describes the model through a casting: gender, apparent age, heritage, build, skin tone, eye color, freckles and moles, hair, facial hair and makeup and look, plus up to two hundred characters of details. Every field has a “?” that explains what it controls, and Heritage and Makeup and look describe the chosen option. Bruxel generates a five-view session (face front, three-quarter, profile, half body and full body) so you can review face and hands carefully.

• The first proposal is generated when the model is created. If it doesn't convince you, you can request up to two new proposals at no extra cost; the previous one is kept until the new one arrives.
• "Accept this model" makes it ready for photos. Until then it doesn't appear on any product page.
• Archiving removes it from the list without deleting the photos you already generated with it. Your organization can have up to 20 active models and create up to five per day.
• The model's five views don't appear in the image library or in the product picker: they are the model's identity, not product photos.

Photos with a model

On the product page, below its images, you'll find the "Photo session". You pick a ready model, one to three real photos of the product as reference and, optionally, a direction of up to a thousand characters ("holding the lipstick next to the face"). You can request one or two proposals per session.

• The better your real photos, the more faithful the packaging comes out. References can only be real photos of that product; a generated photo can't be used as a reference.
• A session usually takes under a minute. You can leave the page: we'll let you know in the bell when it's ready or if it failed.
• Check the hands and the small text on the packaging before saving: that's where mistakes happen most.
• If you sell clothing, pick “Clothing: the model wears it” above the reference photos: the model wears the garment instead of holding it, and fidelity focuses on fabric, print, cut and length. The product page preselects it when the name, the category or a size attribute gives it away, and remembers your last choice for that product. Glasses, bags, jewelry and shoes go under “Product” with a hint like “wearing it”.
• Pick the model's expression for each photo — natural, smiling, laughing, neutral, serious or playful — with the faces above the hint; on Video they appear when the first frame contains the model. The model's five views are generated neutral on purpose: they anchor her identity, and each photo's face is decided when you generate it.

A photo with a model can't be the product's main image. The main image always comes from a real photo; that's what the "Main photo" tab is for.

Clothing: have the model wear it

Above the reference photos there are two options: “Product”, where the model holds or uses what you sell, and “Clothing”, where she wears it. Bruxel preselects the right one when the name, the category or a size attribute gives it away, and remembers your last choice for that product; the decision is always yours.

• In “Clothing”, fidelity focuses on fabric, print, cut, length, neckline and sleeves. If someone else is wearing the garment in your photo, that is fine: only the garment is taken.
• Glasses, bags, jewelry, watches, caps and shoes go under “Product”, with a hint like “wearing it”. They are accessories, not garments, and come out better that way.
• The photo is a full-body shot, in the scene and with the expression you asked for. It takes a little longer than a normal photo because it is built in two steps: first the model in the scene, then the garment on top.
• One garment per photo. For an outfit — a top and trousers, say — generate one photo per garment, or use a real photo where the outfit is already complete.

The model's expression

Every photo picks its expression with the faces above the hint: natural, smiling, laughing, neutral, serious or playful. It opens on “natural”, which is a soft smile.

• The model's five views are generated neutral on purpose: they anchor her identity, which is what makes her the same person across all your photos. Each photo's face is decided when you generate it, not when you create the model.
• On Video, the faces appear when the first frame you picked contains the model. If the frame is a product photo with no people, there is no face to direct.
• In bulk, the expression is picked once and applies to the whole run.

Hairstyle, makeup and wardrobe for each photo

Below the expression icons there are three lists: Hairstyle, Makeup and Wardrobe. They change this photo only, not the model. They open on “As in her model”, which uses what you chose when you created her.

• Her face, her hair length, color and texture, and her build never change: that is her identity. The hairstyle says how that same hair is worn.
• In Clothing there is no wardrobe to choose: your product's garment is the wardrobe. You can still change her hairstyle and makeup.
• Video has no such lists. The clip animates an image that already has its hairstyle and wardrobe; choose them in the photo you will use as the first frame.
• In a batch they are chosen once and apply to the whole run, just like the expression.

Main photo

You pick a single real photo and Bruxel generates its studio version: clean background, no people or hands. It's the only proposal that offers "Use as main photo"; when you use it, the real photo that was main moves to the gallery.

Video

In the "Video" tab of the photo session you pick an image of the product — real, or generated and already saved — as the first frame, and describe the motion with one of the suggested ideas or in your own words. The clip lasts five seconds, vertical for stories or square for the feed.

• It costs 50 credits, reserved when you send and refunded if the video fails, is cancelled or doesn't finish within 24 hours.
• It takes between one and five minutes. You can leave the page: we'll let you know in the bell.
• The video doesn't carry the embedded label the photos do (video formats don't support it yet). In the app it's declared as created with AI all the same and, if it starts from an image with a model, it inherits the synthetic-performer mark.
• For clothing, the same switch on the Video tab changes the motion ideas (turn slowly, walk toward the camera) and keeps the garment unchanged from frame to frame. The path is: photo with the garment on → Save → Video with that image as the first frame.
• You don't pick the video's model in the tab: the clip inherits it from the image it starts from. For a video with your model, first generate a photo with them in "Photos with a model", save it to the product and pick it as the first frame. If you have a ready model and no saved image carries them yet, the tab tells you and offers the shortcut.

Extended video: up to 30 seconds

Next to “Video” there is an “Extended video” tab. It starts the same way, from an image already saved to the product, but the clip lasts between 10 and 30 seconds, can carry audio and is generated by a different engine. It's the tab for an ad; the five-second one is for a detail on a loop.

• You pick the length — 10, 15, 20 or 30 seconds — and the quality. The price goes per second and rises with the quality: each length already shows its total, and the button repeats it before generating. The numbers are in “What it costs”, further down.
• 480p is for checking the idea: the same shot costs half. 720p is the one you publish.
• The audio it generates is ambience and effects. There is voice only if you ask for the AI script and that script writes dialogue, in the language you're using Bruxel in. With “No audio” the clip comes out silent and the script doesn't write dialogue either.
• It takes several minutes, sometimes more than ten. You can leave the page: we'll let you know in the bell when it's ready.
• Your organization gets six extended-video attempts a day, and what counts is the attempt, not the clip: if one fails we refund the credits, but that attempt is already used. When it won't let you make another video today, that's the cap: the tab stops you with a too-many-requests notice and a long wait in seconds. The five-second one keeps its own separate count.
• The finished clip arrives as a proposal, just like a photo: you have to press “Save to product” for it to stay on the product page. And like every proposal, you were charged for generating it: if it comes out fine but you don't like it, the credits don't come back. At this price, it pays to test the idea at 480p.
• The “Product” / “Clothing” switch above applies here too: in Clothing, the motion and the fidelity take care of the garment.

Your model in a long video: the shot-by-shot video

The long-clip engine rejects faces, so the “Put your model in the cuts” checkbox has been removed. For a long video with your model, pick as the first frame a photo they appear in — generate it in “Photos with a model” and save it to the product — and the clip is built shot by shot with an engine that does accept them.

• The first shot starts from your photo, each of the following ones starts from the last frame of the previous one, and the last one ends on your photo: the video flows without jumps and ends on the pose you picked.
• It's charged per second, like the extended video. The panel tells you how many shots it takes and how many credits it costs before you generate.
• You pick the framing as in the short video: vertical for stories or square for the feed. It takes no extra product images and no last frame.
• It takes several minutes: each shot is generated and then they're joined. You can leave the page; we'll let you know in the bell.
• The AI script works here too: it writes one shot per clip, without citing references.

The AI writes the script

Inside the extended video, “Write the script with AI” shows Claude your first frame and gives you back, shot by shot, what will happen in the clip: at which second, with what camera move and what you hear. It costs 3 credits, charged only if the script comes out.

• Before asking for it you pick where it will be posted — that changes the pace and the hook — and, if you like, write in one line what you want shown. That line is optional.
• Don't name the social network there. The filter that blocks asking for real people also reacts to “tiktok” or “instagram” and rejects the request without charging you. That's what the selector above is for.
• For each shot you can rewrite “What happens”; the timings and the camera are fixed. But careful: editing the cards doesn't change the clip. What travels to the engine is the text below, under “See the text the video model receives”: open it and edit it there. When you generate we paste our rules on top — the product's fidelity and that every person is fictional — and those can't be removed.
• Editing costs nothing. Asking for another script charges the 3 credits again. When you come back to the product page, the script of the last clip you did generate is restored: if you write one and leave without generating, that one is lost.
• Ask for the script with the length and the mode already picked: if you then change the length or the quality, the script doesn't rewrite itself and its shots still count the earlier seconds. There are 60 scripts a day per organization, plenty to ask for it again.

3D object: the product can be spun and viewed in your space

The Studio can build a 3D model of your product from its real photos: the customer spins it, zooms in and, from a phone, sees it at full size in their own space with augmented reality. It lives in the «3D object» tab of the photo session, next to the others.

• Pick 1 to 4 real photos of the SAME product, from different angles. The first one you mark is the front: that's the one the generator uses as its main reference. The more angles, the better the result.
• It costs 75 credits per object, no matter how many photos you use. It takes 2 to 10 minutes; you can leave the page and we'll ring the bell when it's ready.
• It takes up about 30 MB of your space: the model and its iPhone version are stored, plus the thumbnail.
• A product has a single 3D object. If you save a new one, the previous one is unlinked and stays in your library.
• You can generate up to 20 3D objects per day.

Save or discard

Proposals show up in the same section, under "Proposals from your sessions", and already live in your library when they arrive. At the top goes what's left to decide — from all your sessions, with a label saying which one each came out of — and what you already saved or discarded folds into "See the earlier ones". Each one offers two actions:

• Save to product (or Use as main photo): links it to the product page. It's still the same library image, not a copy.
• Discard: sends it to the Images trash, where you can restore it for 30 days.
• The cost was for generating. Saving or discarding costs nothing, and discarding doesn't refund credits.

What it costs

The price is shown on the button before generating and charged when the session is sent. If the session fails, is cancelled or rejected, the credits are refunded. Studio spending counts toward your organization's spending cap per period, just like descriptions. Before generating, the panel tells you how much space the photos will take and how much you have left; if they don't fit, you know before anything is charged.

The "Created with AI" label

Every image Bruxel generates is declared as generated. You'll see it in two places, and the file carries it inside:

• In the app: the "Created with AI" badge on the product page, on proposals and in Images, where you can also filter "Generated with AI".
• In the original file: embedded metadata with the digital source type that identifies AI-generated content and, when a model appears, the mark that it contains a synthetic performer. This is the data Amazon and Google require for this kind of content.

Who can do what

• Create, accept, request another proposal or archive a model: administrators and owners. It's a brand decision, like the voice.
• Generating a product's photos, keeping or discarding them: operator role or above. Generating spends credits, and the editor is precisely the role that doesn't spend.

When something doesn't add up

Studio messages always say whether you were charged or not. The most common ones: