The Prompt Composer

The Prompt Composer Icon Prompt Composer is a workspace designed for high-fidelity prompt engineering. It allows you to visually design your layouts, position typography, and configure precise aesthetics.

It is natively optimized for models that support structured JSON prompting (such as Ideogram 4, Flux.2 [Klein], ERNIE-Image, and Lens) and translates your visual layouts into natural language descriptions for standard text-to-image models.

1. Scene Summary & AI Auto-Layout

The primary starting point in the Prompt Composer is describing your overall vision in the Scene Summary. You can then let the Studio Assistant build the initial spatial layout for you.

  1. Describe the Scene: Write a short description in the Scene Summary input (e.g., "A wizard on a mountaintop casting a spell under the night stars, with the word 'MAGIC' floating above").
  2. Click Auto-Layout: Click the Sparkles Icon Auto-Layout button.
  3. Deconstruct & Arrange: The Assistant will analyze your description, set the backdrop, create separate boxes for the wizard and the text element, place them at the correct positions, and fill in detailed prompt instructions for each component.

The Auto-Layout feature utilizes local intelligence. Ensure you have installed the Studio Assistant Model in Settings > Assistant to enable it.

2. The Interactive Layout Canvas

Once the layout is created, you can interactively refine the composition of your image before generating.

  • Draw Custom Boxes: Click and drag anywhere inside the canvas area to create a new layout box.
  • Add Objects & Typography:
    • Object Icon Object: Describe subjects, characters, or props (e.g., "a sleek black coffee mug").
    • Text Icon Text: Specify the exact text content you want to render and its typography style (e.g., text: "MINDCRAFT", style: "bold futuristic neon orange lettering").
  • Position & Resize: Drag elements to reposition them, or grab the corner handles to resize.
  • Keyboard Shortcuts: Select any element and press Delete to instantly remove it.

3. Aesthetics, Style & Custom Palettes

Below the canvas, you can define the overall artistic direction of the entire scene:

Medium & Style Details

Select your creative medium (e.g., photograph, digital illustration, 3D render, watercolor). The available options will dynamically change to match the selected medium:

  • Photographic Medium: Adjust camera settings (e.g., 85mm portrait lens, macro lens close-up) and control depth of field (e.g., shallow depth of field, bokeh background).
  • Artistic Medium: Configure art descriptions (e.g., oil painting with thick brushstrokes) and surface textures (e.g., matte finish, rough paper).

Atmosphere & Lighting

Choose from professional lighting setups (e.g., warm golden hour light, dramatic rim lighting, volumetric light rays) to set the mood of the generation.

Custom Color Palette

Select and edit custom color palettes (up to 5 colors) using an interactive color picker. This maps directly to the model's color generation engine, ensuring your output strictly follows your chosen color scheme.

Not all models support custom color palettes. Please refer to the model documentation for more information.

4. Under the Hood: Positioning Modes

You don't need to manually configure how spatial positioning is sent to the AI. MindCraft Studio automatically handles layout calculations behind the scenes depending on your selected model:

  • Ideogram 4: When Ideogram 4 is selected, the composer automatically outputs raw normalized bounding box coordinates ([ymin, xmin, ymax, xmax]) in the [0, 1000] viewport space. This guarantees pixel-perfect layout control and precise text placement.
  • Other Models: For standard text-to-image models (e.g., Flux.2 [Klein], Z-Image Turbo, ERNIE-Image, Lens), the composer dynamically translates your visual canvas layout into qualitative natural language position terms (e.g., "spanning top horizontally", "top center", "bottom left") so the model can easily interpret your composition.

5. Applying & The Bidirectional Workflow

Once you are done composing, you have two options to apply your prompt:

  • Compose Icon Compose & Apply: Compiles your visual canvas, background, custom styles, and positions into a beautifully structured natural language paragraph, then applies it to the prompt field.
  • JSON Icon Apply as JSON: Compiles all settings into a structured JSON string. This is the optimal format for Ideogram 4, delivering maximum text rendering accuracy and compositional fidelity. Other models support JSON prompting (Flux.2 [Klein], ERNIE-Image, Lens) can also benefit from it.

Bidirectional Editing

If you ever want to tweak a previous creation, simply copy and paste its structured JSON prompt back into the prompt editor. The Prompt Composer will automatically parse the code, rebuild the visual boxes on the canvas, restore your color palettes, and populate all advanced settings, allowing you to resume editing right where you left off.

Next Up

Model Prompt Guides

Comprehensive prompt guides for every supported model: GPT-Image-2, Ideogram 4, Flux.2 [Klein], Z-Image Turbo, and Google Nano Banana.

Continue