Accurate Syntax, Parameter Rules and Multi-Platform Adaptation
Generative AI models do not share a universal syntax. Pasting Midjourney command flags into DALL-E produces meaningless errors, while feeding natural-language paragraphs into older Stable Diffusion models can dilute key keyword weights. This guide outlines the exact documented parameter rules for the four leading image generation platforms.
Keep a portable base prompt
A natural-language visual brief is a sensible starting point across platforms, but outputs will differ by model and interface. Do not assume that a prompt gives identical results in Midjourney, Stable Diffusion, DALL·E or Leonardo AI.
Midjourney parameters are not universal
Our Midjourney format appends --ar with the chosen aspect ratio. Do not paste this parameter into other platforms expecting it to control dimensions. Check the current official interface documentation for supported ratios and any other parameters you intend to use.
Use exclusions only where supported
Stable Diffusion workflows often provide a separate negative field; availability and effect depend on the model and workflow. Leonardo support also varies by model. For interfaces without a negative field, describe the desired scene positively rather than relying on a copied exclusion list.
Record the settings you actually used
Keep model version, dimensions and other supported settings with your prompt when comparing results. A ratio describes proportions, not resolution. Choose the closest supported size; do not assume typing “4K” produces a particular pixel count.
Platform Rules & Parameter Interpretation
Midjourney (v6 / v6.1)
Prompting Style: Short to medium natural-language descriptive phrases with high sensitivity to artistic, photographic, and lighting adjectives.
Syntax & Parameters: Uses inline double-hyphen parameter flags at the end of the prompt: `--ar [width:height]`, `--style raw` (reduces default aesthetic bias), `--stylize [0-1000]`, `--weird [0-3000]`.
Negative Prompt Handling: Supports `--no [unwanted items]`. However, positive descriptive framing is almost always more reliable than excessive negative lists.
Stable Diffusion / SDXL
Prompting Style: Responds exceptionally well to structured, comma-delimited tokens grouped by importance: subject, environment, lens, lighting, styling.
Syntax & Parameters: Dimensions are set in the generator interface (e.g. 1024x1024, 1344x768). Supports token weighting syntax like `(term:1.2)` or `[term]` depending on UI.
Negative Prompt Handling: Has a dedicated Negative Prompt input field. Highly effective for filtering out common deformities: "deformed hands, extra fingers, plastic skin, cartoon, blurry".
DALL·E 3 (OpenAI)
Prompting Style: Strictly conversational, natural-language descriptive prose. DALL-E uses an internal language model to expand and interpret your brief.
Syntax & Parameters: Does NOT support parameter flags (no `--ar`, no weights, no tags). Dimensions are selected via interface buttons (square, portrait, landscape).
Negative Prompt Handling: Does NOT have a negative prompt field. Exclusions must be phrased positively or directly: "A clean modern room with no furniture or clutter."
Leonardo AI
Prompting Style: Hybrid workflow supporting both natural language and tag tokens, powered by fine-tuned diffusion checkpoints (e.g. Leonardo Phoenix, Kino XL).
Syntax & Parameters: Preset styles (Cinematic, Photography, 3D Art) act as embedded negative and positive style prompts. Aspect ratio is controlled via sidebar sliders.
Negative Prompt Handling: Features an explicit Negative Prompt toggle. Effective for suppressing cartoon rendering when generating photorealistic scenes.
The Portable Prompt Rule
Keep your core visual prompt clean, descriptive, and platform-neutral. When switching platforms, copy the natural-language description as your foundation, then append platform-specific flags (`--ar` for Midjourney) or populate interface fields (Negative Prompt for SDXL) in their respective tools.
