Prompt studio
Start with a sentence.
Write the picture as you imagine it.
Mark the parts you want to define.
01 Your prompt
Write freely, then select any words and give them a role. For Ideogram JSON, label each distinct subject and the scene for a reliable element layout.
Select text with your mouse, keyboard or touch.
Composition, text & references Optional
Auto-detection responds to explicit live-action, photographic or film-still wording; “cinematic” alone does not activate it. Explicit 3D render, cartoon, illustration or painting intent takes precedence in Auto mode. Locks are carried into Preserve for edits.
MING / MOSS prompt format Model-specific
Checked 11 November 2026. Select the real checkpoint and workflow: native generation, documented colour/depth control, or semantic layer decomposition. Prompt text, compiled layout JSON and runtime controls are kept separate.
The JSON below is the complete generation prompt. Order coherent layers from background to foreground. Coordinates are normalised centre x/y and width/height; use each exact text string once in its description. Structured generation JSON is not an edit request body.
Supply one existing flattened design image to Design-Layer. Write one semantic RGBA output layer per line, front to back, with the background last. Put text in front of its supporting card or panel. Count is calculated from this plan. Only this plan is copied; the main paragraph and styles are not used.
Set the generation canvas below; it overrides the aspect-ratio dropdown for layouts and must stay unchanged after compilation. Author boxes use [left, top, right, bottom] pixels from the top-left. Copy validates the official draft schema and emits the model’s one-line centre-origin yxyx JSON. The draft is the complete prompt source; other prompt fields are not merged. English captions are recommended; exact visible wording can include other languages.
FLUX.3 layout boxes Optional
Hosted prompt format only: write element ids such as <subject_1> in your paragraph, then add JSON rows below. Use [top, left, bottom, right] integers from 0 to 1000. Leave empty for ordinary prompting. For edits, use from: ref_image_0 with source/target boxes to move or anchor; use from: null and src_bbox: null to add or replace, or tgt_bbox: null to remove. Local Python boxes use a separate 0–1 contract (including from: target for additions); do not paste these hosted rows into that argument.
Keep the edit precise
Use your paragraph for the change. State the details that should stay the same.
Camera angle & perspective Optional
Choose one viewpoint when it helps show the action, scale or spatial relationship. Keep a neutral view for reference sheets, accurate products and precise layouts. For edits, ask an editing model to reframe the existing image.
The edit keeps the subject and setting, while reconstructing surfaces newly visible from the chosen angle. Add any specific identity or background details under Preserve.
02 Visual direction
Choose a main style, then add a look or mood if it helps.