Generate image from text

The Generate image from text mode (Generate) creates images from scratch using only a prompt. It is the right workflow when you do not want (or cannot) start from an existing photo: you need a concept, a visual, an illustration, or a scene that exists only “in your head”.

In this mode the main controls are:

• the prompt (what you want to see),
• the style (realistic, cinematic, illustrated…),
• and, if supported by the selected model, format and resolution (ratio and output size).

Use Generate image from text if:

• you need concept art, moodboards, key visuals;
• you want to generate a scene, a product, a character, or an illustration;
• you want many quick variations of the same idea;
• you want to start from a description and then iterate.

• If you already have a photo and want to edit it: use Single image edit.
• If you want to change only the background of a photo: use Background edit (fast).
• If you want to merge multiple photos: use Compose multiple images or Seamless Blend.
• If you want to turn a sketch into an image: use Sketch to image.

• A descriptive prompt.
• No input images.
• (Optional) Prompt Builder to write more structured prompts.
• (Optional) Ratio and resolution, if the model supports it.

When you start generation, the app sends to the model:

• the prompt text;
• any format hints (ratio) and size (resolution) if available;
• any constraints/style you entered (or derived from the Prompt Builder).

The model returns a new image.

In this mode you typically see:

• Mode: workflow selector (choose Generate image from text).
• Prompt: main input field.
• Prompt Builder: guided fields (style, light, colors, mood, camera/lens…).
• Ratio and Resolution (if available): control format and output size.
• Generate: start creation.
• Generate variants (2-4): alternatives of the same idea.
• Use as input: bring the output into Edit for image-based iteration.
• Image analysis (if available): describes an output to turn it into a prompt.

1. Write a simple one-line prompt to lock the concept.
2. Generate and review.
3. Generate 2-4 variants and pick the best.
4. Refine the prompt (add details, constraints, style).
5. If you want control over specific areas, use Use as input -> Single image edit and work with markup.

In the “Mode” selector, choose Generate image from text.

Write as if you are describing a photo or a cinematic frame:

• main subject,
• environment,
• action/pose,
• light,
• style,
• important details.

Example: “Realistic photographic portrait of a woman in studio, soft softbox light, neutral gray background, 85mm, natural skin, high definition.”

If you want fewer random attempts, use the Prompt Builder to compose:

• style (cinematic, watercolor, photorealistic…),
• lighting,
• colors,
• mood,
• details,
• camera/lens.

Tip: use the Builder for structure and add 1-2 sentences in the prompt for unique details.

If the model supports controls:

• choose ratio (1:1, 16:9, 9:16…);
• choose resolution (Auto/1K/2K/4K).

If you do not see these controls or they have no effect, the selected model likely does not support imageConfig.

Click Generate and evaluate:

• consistency with the prompt,
• detail quality (hands, text, patterns),
• style and light.

• Use Generate variants (2-4) to explore alternatives.
• If an output is almost perfect but missing something:
  • edit the prompt (add/limit),
  • or use Use as input and refine in Edit.

1. Subject: who/what.
2. Environment: where.
3. Light: what light and from where.
4. Style: realistic photo, illustration, cinematic, etc.
5. Details: materials, textures, palette.
6. Constraints/negative: what to avoid.

“[SUBJECT], [ENVIRONMENT]. Lighting [LIGHT]. Realistic photographic style, high definition, natural textures, believable colors. Avoid text/watermark and artifacts.”

“Illustration of [SUBJECT] in [ENVIRONMENT], [WATERCOLOR/OIL/LINE ART] style, palette [COLORS], composition [RULE], clean details.”

“Studio product photo of [OBJECT], background [WHITE/GRAY], softbox light, realistic contact shadow, sharp details, no text.”

If needed, add a final line: “Avoid: text, watermark, deformed hands, artifacts, oversharpen, plastic skin.”

Great for:

• changing micro details without rewriting everything;
• finding a better composition;
• getting a cleaner render.

The most effective way to go from a “good idea” to a “final image”:

• generate a solid base,
• then bring it into Single image edit to control specific areas with markup.

• Text generation is probabilistic: the same prompt can produce different results.
• Perfect text and lettering are not guaranteed: if you need precise text, add it in post production.
• If the model does not return an image, it may be a policy block: simplify or change the request.

• Make the prompt more specific and shorter (less ambiguity).
• Put the most important information first (subject and scene).
• Generate variants and pick the closest, then refine.

• Add constraints: “natural hands”, “symmetrical eyes”, “clean pattern”.
• Generate variants and then refine with Edit (Use as input).

• Explicitly state the style and repeat 1-2 key characteristics (e.g. “photorealistic, natural skin texture”).
• Use the Prompt Builder to lock the style.

• Avoid requests involving sensitive or protected content.
• Try a more neutral, gradual prompt.