Follow it once
and you have made the thing

Example one

Three style-matched promo images for a coffee shop

A pour-over close-up, the room, a takeaway cup. All three should read as one shoot: same palette, same light, same grain.

All four images use Midjourney V7 at 10 credits each, 40 credits in total.

Steps

1. 01

   Create a Project and pick Canvas

   Click "New project" on the left, give it a name, choose Canvas.

   The dialog settles four things at once: which Studio it belongs to, what the Project is called, which kind of Space it opens with, and the slug in its address. Canvas is "Infinite canvas + nodes", Document is "Rich text + collaboration", Timeline is marked "Not available" and cannot be picked.

   The slug is not filled in for you, so type one; leaving it empty also works.

   

2. 02

   You land on an empty canvas

   The middle of the canvas reads "The canvas is empty" and "Drag assets onto the canvas, or pick a node type from the left menu to create one."

   The vertical strip on the left is the create menu; the cluster bottom-right is the viewport toolbar (undo/redo, zoom, fit to window, grid snap, minimap). The whole left column is the Agent panel.

   

3. 03

   Create an image node from the node library

   Click "Node library" at the top of the create menu.

   There are four node types: text, image, audio, video. This example makes images, so pick Image.

   

4. 04

   The empty node appears

   The node reads "Double-click to upload" and "right-click to generate & more" — two routes: upload what you already have, or right-click to have it generated.

   

5. 05

   Right-click the node and choose Generate

   Alongside "Generate" and "Upload" the menu carries "Reset to empty image", "History", plus copy, rename, lock and delete. Tools is greyed out.

   

6. 06

   The generate panel

   The panel hangs below the node. Top to bottom:

   Where	What it is
   Three buttons, top left	The reference slots: Reference, Focus, Style
   The large box	The prompt. Its placeholder reads "Describe the image you want to generate, @ to add a reference"
   Bottom left to right	Mode (Text to Image), model (Midjourney V7), ratio and resolution (1:1 · 2k)
   Bottom right	Web search (greyed out), what this run costs (10 credits), the generate button

   Switching model changes the whole panel, not just the price — resolution tiers, which reference slots work, extra controls. This example stays on Midjourney V7 throughout because its style slot is available.

   

7. 07

   Write the first prompt

   The first one reads:

   The two images after this one follow its style, so spell out the light and the texture here — “morning light from a side window”, “warm tones”, “film grain” are what set the look of the whole set.

   Close-up of a pour-over coffee, dark wooden table, morning light from a side window, steam rising through the light, blurred background, warm tones, film grain

   

8. 08

   Hit generate and the node starts loading

   The node turns into a grey gradient and a green label appears top-left naming whoever is occupying it. In a shared Project that label tells everyone else the node is in use.

   Midjourney V7 takes a while. Across this example's four images most came back in around a minute, the slowest well past two and a half. You can build the next node while you wait.

   

9. 09

   The first image arrives

   The node is marked 1024×1024 top-right. This image is the style source for the rest of the set.

   

10. 10

    Create a second image node — it lands on top of the first

    Node library, Image, same as before.

    A new node always lands dead centre of the canvas, so from the second one on it will cover what is already there. Nothing is wrong; drag it off.

    

11. 11

    Drag the new node aside

    Drag the node into open space. With both visible you can point at the first one when you pick the style reference.

    

12. 12

    Right-click the new node, Generate, then click Style

    Click "Style" in the top-left of the panel. The canvas enters pick mode: a bar appears at the top reading "Select a style reference from the canvas", with a locate button and "Exit" beside it. Every image node on the canvas is now clickable.

    No wiring needed. Just click the image on the canvas.

    

13. 13

    Click the first image; the slot becomes its thumbnail

    The "Style" button turns into a thumbnail of the first image with an × to clear it. Pick mode exits on its own.

    

14. 14

    Write the second prompt

    The second one reads:

    The subject changed; the style wording carried over from the first prompt. Together with the thumbnail in the slot, that is two things pulling the same way.

    Interior of a coffee shop, wooden counter and pendant lights, a few customers by the window, morning light slanting in, warm tones, film grain

    

15. 15

    The second image, next to the first

    The subjects have nothing in common — one is a tabletop close-up, one is a room — but the warm light, the wood and the shallow focus do.

    

16. 16

    The third image, the same way

    Another node, dragged clear, right-click generate, Style, click the first image. The prompt becomes:

    Put the three together and they read as one shop, one hour, one camera.

    Still life close-up of a takeaway coffee cup on a wooden surface, coffee beans scattered beside it, side backlight, warm tones, film grain

    

17. 17

    The control: what happens with no style reference

    One more node, no style reference, the exact same prompt as the third image.

    

    Bottom right and bottom left are the same prompt, twice: the right one had a style reference, the left one did not.

    Aspect	With a style reference	Without
    Background	Warm brown, thrown out of focus	Bright cool planks, every detail sharp
    Light	Side backlight, the first image's direction	Flat and even, from nowhere in particular
    Overall	Reads as one of the set	A clean product shot on its own

    The prompt spelled out “side backlight”, “warm tones”, “film grain”, and the bottom-left image went its own way anyway. Words cannot hold a style; one reference image can.

Things that came up along the way

What this example leaves out

@ references, the Focus and Reference slots, image-to-video, Document, the Agent panel. Those belong to later examples — one example, one thing. @ references are the product's other core interaction and deserve an example of their own.

Models

What you can pick from

You choose a mode first — text to image, image to video, transcribe — and the picker then offers the models that serve it. Credits are charged per run; the times are the config's worst case, not an average.

A deployment only offers the models it holds an API key for, so what you see in the product can be a subset of this list.

Chosen for you

Three places we pick the model

Everywhere else the model is yours to choose. These three run on one we set, and your text reaches it the same way it reaches any other provider.

• Nano Banana turns your description into a structured prompt before it draws — the Edit variants skip this deepseek/deepseek-chat
• The editor's ten text tools: polish, expand, summarise, translate, rewrite, continue, generate, character, storyboard, script google/gemini-2.5-flash
• The assistant's replies, and the summaries it writes into memory deepseek/deepseek-v4-pro

If you want more