Generate an image

1. 1

   Open Image → Create and describe the picture. @mention a character, an element or a recent render to bring it into the prompt; Add character and Add context elements, above the box, open the same list.

2. 2

   In Settings, pick a model card (each one shows its price and what it is good at) and a format (1:1, 3:4, 4:3, 9:16 or 16:9).

3. 3

   Choose 1 to 4 takes with the − and + under the prompt. The button shows the total; click Generate. An empty box is not an error to hunt for: the button puts the cursor in it and says "Describe the picture first."

4. 4

   Each Generate adds a row to the feed under the composer, newest on top, with its takes developing in place. Hover a take to Animate, Edit, Download or delete it; click it to open it in the viewer. "Clear this temporary feed" empties the page; the renders stay in History and the Library.

The controls

• Negative prompt (on the models that read one): what to keep out of the picture. It is never checked against the content policy, so "child" there is the right way to exclude one.

• Prompt strength: how literally the model follows your words, from Loose to Strict. Move it only when a render ignores you (raise) or looks stiff (lower).

• Enhance: no credits. On an empty box it invents a prompt on every account; rewriting your draft as a fuller description needs a paid plan. Mentions survive the rewrite.

• Save to: the Library folder new renders go to, shared by every tool.

• Style, beside Add character: six looks (iPhone photo, Photorealistic, Anime, 35mm film, Cinematic, Flash editorial). A pick writes its words at the end of your prompt, where you can edit them; another pick swaps them, and "Remove the style" takes them out. There are no image templates any more: example prompts rotate in the empty box instead.

Text and logos

Diffusion models mangle lettering and eroq has no text layer. Leave room for the type and add it in a design tool, or use the Photoshop or Figma plugin and keep the whole workflow in one place.