Create an image makes an image from the scene you write. Text alone is enough; you can add references to it: up to four subjects (characters, animals, objects) and one location, connected to the generator. It is the image counterpart of Create a video: same cards, same @, same library.

Since 0.9.0, this screen replaces the former "Generate an image" and "Multi-Ref Image". Free LoRAs (a LoRA file picked from a list, outside any character) no longer exist: a LoRA goes through its character.

Connect the references

Click Add a subject (up to four) or Add a location (one only). The window has two modes:
- Import: drop an image, click Browse, or take it from a tab. Gallery takes an image you already made; Characters, Animals and Objects (or Locations for the location) take it from your Assets. Each tab starts with a card to create the element (creating a character needs a licence).
- Generate: describe the image you want, with one of your trained characters if you like, and the app suggests it. Online, each suggestion costs the price of one image; on your computer, it is free.
An image from the gallery, dropped or generated, is connected with Add to generator. An element from your Assets is connected with one click on its card.
For a character, then choose what stands for it (see below): its LoRA, or an image from one of its character sheets.
Each card has a switch: a card turned off stays in place but is not sent to the render.

Objects, animals and locations are also created in Assets > Elements, with up to six images, a title and a description.

A character: an image of its sheet, or its LoRA

When you connect a character, the window shows its character sheets and, if it has one, its LoRA at the top.

An image of its character sheet. Click the image that shows it best for this scene. Create an image takes a single image per character (Create a video takes the whole sheet). Choose a view where its face shows: from behind, it would come out as a stranger. All about sheets in Character sheets.

A character with no character sheet starts from its profile photo alone, as it is, and the likeness is weaker. Its card shows No character sheet, and a click on Generate opens the Limited likeness window: Create the character sheet or Generate anyway.

Its LoRA. The Character LoRA tile only appears for a character that has an image LoRA (see Train a LoRA). It is the most faithful identity. Alone in the image, the character comes in with its LoRA; with others, the app first composes the scene, then goes over each one's part of the image again with its LoRA. On its card, the LoRA badge opens the choice again.

The cards: Strength and LoRA pass

Strength: at 0, the reference is ignored. For a character played by its LoRA, it is the strength of its LoRA: 1 by default, 1.5 at most.
LoRA pass (30 to 90 %, 55 % by default): on a character played by its LoRA, when the image has several subjects. It is the share of its part of the image that its LoRA repaints. Higher, it is more faithful to its LoRA (hair, outfit) but the background around it moves. Lower, the scene stays and the LoRA mostly corrects the face. Alone in the image, a character has no such slider: its LoRA is in the whole image.

Write the scene

Call each reference by @ and its name, where you want it in the sentence: "@Lea hands a cup to @Lucas on the terrace". A card that is on is sent even if its @ is not in the text; the @ says where it goes in the sentence.
Each @ can be clicked. A small window shows what it sends to the model and lets you change it, for example to change an outfit. It also shows its place in the image and the exact sentence sent for it.
Each one's place follows the order of the cards. With two subjects, the first is on the left and the second on the right; with three, left, center and right; with four, from left to right. To swap two characters, connect them again in the other order. Exception: a single character played by its LoRA, with objects or animals, has no set place, so that the written scene is kept ("@Lea leans against @Car").
An image is one moment: a pose, an expression, a gesture. A series of actions does not show in one image.
A character that is not human (a creature, a dragon): say what it is in the scene.

+ Style writes the words of the chosen style at the end of the scene, before your eyes: you can change or delete them. + Shot framing inserts a framing (close-up, wide shot...) where the cursor is.

The AI button rewrites your idea into a scene ready for a still image and keeps the @ as they are, without describing what the characters look like, since their references already bring it. It works online with no credits, or offline on your computer once the local assistant is installed (a one-time download, offered at the first click or from Settings > Components). Back to the original text undoes the rewrite.

What the app does with your cards

You have nothing to choose: the app takes the right method from the cards that are on.

Your cards	What the app does	Number of passes
No card (text alone)	An ordinary image, from your scene.	1
Only characters played by their LoRA	Alone, the character comes into the scene with its LoRA. With others, it composes the scene, then each LoRA goes over its character's part again. The location is described by its text (its description, or else its name), not by its image.	1 for one character; 1 + the number of characters beyond that
Only references by image (people, animals, objects, location), 4 at most with the location	The images themselves go into the render: the real person, the real object, the real place. Beyond two, each extra subject is put in its place by one more pass.	The number of references minus 1 (at least 1)
Both mixed	It composes the scene, each LoRA goes over its character again, then each subject by image is put in its place. The location is described by its text. A single character by LoRA with objects or animals: its LoRA is in the whole image, with no separate pass.	1 + the number of subjects; with a single character by LoRA and only objects or animals by image: 1 + the number of objects and animals

Quality, format and price

Quality: Small (768), Standard (1024) or High (2K), chosen here and nowhere else (the "Generation size" of Settings is gone since 0.9.0). 2K is native, rendered large, not enlarged afterwards. It is for licences only.
Format: Portrait, Square or Landscape; the exact resolution shows underneath.
Images per run: 1, 2, 4 or 8.
Seed: automatic by default, a new one for each render. Fix it with the die to make the same image again while changing only one thing at a time (see seed).

Online, each pass costs the price of one image: the price of a render is therefore the price of one image, multiplied by the number of passes (table above) and by the number of images asked for. It shows next to Generate before the render. On your computer, it is free.

In the demo, Create an image can be tried with the free images of the trial, in Standard (1024) at most. Lea and Lucas come with their full character sheets.

On your computer or online

The Local / Cloud GPU switch decides where the render goes. On Local, nothing goes online. On Windows, Create an image renders on your own graphics card:

text alone;
characters played by their LoRA, if the LoRA file is on your PC;
up to four references by image, if edit by instruction is installed (Settings > Components);
a mix of both, with those two conditions.

Otherwise, a message says what is missing and offers Switch to Cloud GPU. On Mac, text alone renders on your Mac; with references, Create an image renders online for now.

Reuse an image

Once the image is done, the preview offers Edit the prompt, Regenerate (a new seed, even if you had fixed one) and Show file. With several images, move from one to the next in the preview.

In the Gallery (Images tab), Reuse opens Create an image again with its references connected, its text, its style, its quality, its format and its seed. This also works for images made before 0.9.0 in "Generate an image" or "Multi-Ref Image": the character comes back through its LoRA, with its strength; a former free LoRA is left aside. If a reference is no longer in your library, a message says so: connect it again by hand.

To change an existing image (change a detail, paint a zone), use the separate Inpainting screen.

For a clip with the same references, see Create a video.

Create an image: from text alone, or with your characters, objects and locations as references