Skip to main content
Marsell makes pictures in your look: your products kept exact, your colours, your kind of shot. Ask for them on Make, make them from a product, or make the ones the week’s plan needs. Each picture shows its price before you press, and the pictures land in Library in a minute or two.

Make pictures

1

Pick Picture

In Create, on Make, pick Picture under “What do you want to make?”.
2

Say what it's about

Write it in one sentence under “What’s it about?”, for example “A flat lay of the two tees”, or tap one of the ideas under the box.
3

Choose the format and shape

Under How it’s made, pick the format: for posts, Post picture, Carousel or Cover or thumbnail; to sell, Product shots or Ad set (one ad in every size it runs in). Then the shape, such as 4:5 for a post, and how many.
4

Choose what goes in it (optional)

Products, characters (one chip for each look of each), places, your look and reference pictures come in as chips from your Elements. Take any out, swap it, or add a picture of your own or a link.
5

Check and make

Under Models, Auto picks the picture model, or pick one yourself: each says about how many tokens a picture costs. Press Make it, with the price on the button. Make writes the prompts in your look and makes them in one press.
To read and change the prompts before anything is made, press Lay it out on canvas first: it writes the prompts and lays out the parts, and nothing is made until you press Make. Open an empty board is free. If Marsell hasn’t read your look yet, it reads it once first, at the price shown. A carousel’s slides are made as pictures with no words, and the words are set over them in code, so they are always spelled right. You can also ask for a picture in the chat. Once it’s made, Marsell asks whether and where to post it, and offers to run it as an ad only when that makes sense or you ask.

Pictures of a product

In every picture, on the Elements tab, lists your products as they’re sold on your site, and the ones you added yourself on Brand, under What you offer. Press Make with the number on a product’s card to make pictures of it straight away. The product is kept exactly as photographed. Tapping a product under Elements sets up pictures of it on Make instead.

Pictures of real places

A picture of a real place or thing is made from real photographs of it, and Marsell checks the picture it made against them. It looks in this order:
  1. Your own posts on your connected accounts and your own pictures.
  2. For a place, your own YouTube films: Marsell picks the films that may show it from their titles and descriptions, and Gemini 4 Argon watches each one once and lists the places in view and when, by the names on signs or in the narration and the names it knows them by. A place is matched against all the places a film shows at once, so a mosque and the madrasa named after the same founder are never taken for each other. Marsell can’t take frames from YouTube itself, so your own AI is asked, as homework, for the frames at those moments of your film. While they come, the steps below still run, and once a frame comes back it’s used before any other photo of that place.
  3. Wikimedia Commons and Wikidata, searched by the place’s own names: its plain name and the other names it’s known by. A place that is one part of a larger site, such as the theatre of an ancient city, is also looked for among that site’s photos.
  4. For a place or an artwork, the web: an agent searches for pages that show that very place and brings back their photos.
  5. When none of these finds a photograph it can trust, Marsell doesn’t invent one: the picture asks you, or your own AI, for a photo instead.
Marsell’s picture reader looks at every photo before it’s used, and turns down one that shows the wrong place. A film watch and the web search are charged from your tokens at what they cost (a 50-minute film is about 1,800 tokens, once, since Argon moves through a long film and looks only at what it needs; later places in the same film reuse what it saw). A photo from someone else’s site is only a private reference for making the new picture. It’s never posted, never shown on the Places shelf, and How it was made doesn’t show it: it shows a globe, “A photo on the web” and a link to the page it came from. A frame of your film shows as “From your film” with its title and the moment, and opens the film there. Photos most likely to show the place are looked at first: files Wikimedia Commons marks as showing it, then its own picture and category, then search results, and photographs before maps or scans. Search results are sorted by their titles first, so a photo of another building, a coin or a map is never looked at. For one picture, each step of the search (your own material, Wikimedia Commons, the web) asks the picture reader at most eight times before the next step takes over, on every plan; a search that gathers every photograph of a place for a film has no such count. One place’s whole search, from its names to the web, stays within about 1% of your plan’s monthly tokens (2,000 on Solo), so a place with no clear photo stops early. When it stops, it says how many of the likeliest photos it looked at, and the photo is asked of you. Watching your own films is counted apart, once per film. Each place is searched once per project, and later pictures of it reuse what was found. Your own posts’ pictures are read once and kept, so a new place is matched against them without reading them again. Marsell asks you once per place, “Send a photo of” it, however many pictures wait for it. A post waiting for that photo keeps its place and says “Waits for your photo of” the place, and new pictures leave that place out until a photo comes. Every 10 minutes Marsell looks again for free: when a photo of the place turns up in your own pictures or posts, the waiting pictures are made from it. Look again on a waiting picture asks Wikimedia Commons and the web once more and judges only photos it hasn’t judged before. If your own AI gets homework, it is asked to find the photos in your own files too.

Reference sheets

On the Elements tab, a character can get Make a reference sheet and a place Make a place sheet: one picture of them from every angle, made from their own pictures. Before it starts, you see What it makes, what it is Made from, Where it goes and its Price. A sheet is kept as their reference for later pictures and films, and never posted. Marsell makes no sheet of a real person.

This week’s pictures

Marsell plans the pictures your project needs on its own: once when it’s set up, then every week. In a new project’s first week, the pictures the plan needs are made straight away, so nothing waits on an empty card. They start once your brand’s own pictures and one picture of each product are read; a product’s other pictures are read after that. Each planned picture has a purpose and a place to go, such as a post on a given day or an ad, and This week’s pictures in Library shows who the week’s pictures are for. How far the pictures go is up to you, under Pictures made: in setup, and later on Brand, at the end of The look. Settings, under What Marsell does on its own, shows what you picked and Open Brand takes you there.
  • On brand (picked to start): each picture fits your business and the post it’s for, in your own look. Nothing is drawn at random for it: a picture nobody asked for in words varies only in how it’s lit and framed. Your characters appear where a picture is about them, and whenever a picture shows one of them, under any name, its pictures go with it, so it’s drawn exactly as pictured.
  • A playful twist: on brand, with one surprising touch in each picture nobody asked for in words.
  • Wild ideas: a picture nobody asked for in words takes a random era, genre or premise, for a brand that wants to surprise.
Whichever you pick, a picture asked for in words, by you, your board, the plan or Marsell’s chat (in the app, on Telegram or on Slack), follows those words and draws nothing at random, and words you put in quotes appear exactly in what the picture is made from. An ask for a joke keeps your pick: on brand, the joke is made in your own look. All three cost the same. Each card shows the day (or Any day), where it goes, and where it stands: Planned, Making, Ready to keep, Didn’t come out, Kept or Dropped. Open a card to read the idea, Why now, The shot, Made as, Its caption and How it went. One that didn’t come out says why.
  • Make it, with its price, makes one picture; Make it again retries one that didn’t come out. When the picture model runs out of time or is briefly down, Marsell tries once more on its own about 10 minutes later, and only a second miss shows on the picture. While pictures are being made, the top bar says Making pictures, What I’m doing lists them under what runs now, and their piece on the plan reads In progress.
  • Drop takes one out of the week; Bring it back puts it back.
  • Make the week’s pictures, with the number and total price, makes all that are left.
  • Plan the week again rewrites the week’s pictures.
Some planned pictures need a real picture from you, such as a screenshot of your product at work or a team photo. Marsell doesn’t invent these: the card says Needs a real picture: and what. Pick Use this screen from your site or Use one of your screens where Marsell has them, or Send the picture (PNG, JPEG or WebP). For a real place, the card says where Marsell looked, and Look again searches once more.

Keep the good ones

A finished set waits for you under Needs you in Library, and as a card on Home. Keep all keeps the set; One by one opens it in your review deck. A set of one picture asks “Keep the picture for” its title, “It’s ready.”, with Keep it. Keep a picture from its tile, or Keep it when it’s open. Not this one takes a picture out of the Library. Say what’s wrong (Looks fake, Not our look, Wrong product, Odd details, or your own words) or Just take it out. Marsell won’t make pictures like it again. A kept picture goes where it was planned: a draft post for its day and channel, or an ad picture. Marsell learns from your first answers on each channel, then posts on its own. See Posting and the calendar.

Take a picture back or delete it

Open a picture, and at its foot:
  • Don’t use it, on a kept picture: it comes off every post that hasn’t gone out yet and stays in the Library, so you can keep it again.
  • Delete asks once, saying which posts it comes off, then puts it in the bin. Posts already out keep it.
To delete several, press Select in Library, pick them, then Delete. A post that loses its picture leaves the schedule, so it never goes out bare: open the post and Pick another (its other versions, or a picture you kept that no other post uses) or Make another, at its price. Nothing is made on its own. A picture kept for an ad can’t be taken back or deleted here: pause or stop its ad on Sell, under Ads. The Bin, at the end of Library while something is in it, keeps what you deleted for 30 days, with Bring it back. A picture you delete teaches Marsell what turning it down does. See Create.

The look

The look, “Every picture and film is made in it”, is on the Elements tab. It is what Marsell read from your site and your pictures:
  • a summary, Your logo · in every picture, your colours (click one to copy its hex code) and the words for your aesthetic;
  • The motifs, your type, your reference pictures and what never to show.
Posts you already published set the house look too. When your own AI sets you up, it can add them as posts, and Marsell’s new pictures share their finish, light and palette, but never their subject or their words. See Set up a business with your AI. Before the first read, the section offers Read the look: it takes a minute, and “One read, kept for every picture and film after”. Until then, every picture still carries your colours, your type and your logo. Read it again, with its price, has Marsell look at your site and pictures afresh. On Make, Not right? next to your look lets you correct it. To change the look in your own words, or your colours, type and logo, use Brand. You can also tell Marsell in the chat what’s wrong with it.

Which model makes them

Pictures are made by an image model, GPT Image 2.6 Sunburst first. Where more than one model is on, Models on Make lets you choose, and each picture’s price depends on the model. Besides Marsell’s own picks, you can choose Seedream 5.0 Flash, Qwen Image 3 Pro, Grok Imagine 2.0, FLUX.2 Flex and Ideogram 4.5. Your product pictures and logo go to the model as references, so the product in the frame is the real one. Each picture is made by whichever service sells its model for the least (kie.ai, Sume, OpenRouter, fal or Higgsfield), and its details say which one made it. GPT Image 2.6 comes from kie.ai, Sume or OpenRouter; when none of them can make a picture, Higgsfield makes it with GPT Image 2.5 for the same price. A picture keeps the shape it was asked for: when none of those services makes that shape with its model, it is made at the nearest shape and widened to it, never cropped.

What pictures cost

Each picture costs the model’s list price for its shape and the pictures it is sent, to the token (a GPT Image picture is priced from an estimate of what it uses, so it carries a 10% margin), plus a look at the finished picture that counts at least 225 tokens. A picture widened to its shape adds 400 tokens. A picture of a real place adds three more looks: one that picks the photograph whose view fits the picture and describes the place as that photograph shows it (a silhouette at dusk stays a silhouette), and rewrites a line or a title that names a view the photograph doesn’t show, one at its photographs and one at the comparison. The price on the button is the price charged. The model’s price is held while the picture is made and charged only when it comes out. If the model refuses the picture or can’t make it, that price isn’t charged. Only a picture whose answer was lost, after the model may have made it, is charged as made. Writing new prompts adds the writer’s own cost. Pictures Marsell makes on its own for the plan, a post or an ad come out of the same tokens. See Marsell plans and tokens.

Films

Write and direct a film.

Brand

Your logo, colours, type and look.