Prompt Writing Tips for Magicdoor Image Models

Magicdoor does not currently list Stable Diffusion in its managed image lineup. This guide focuses on prompt-writing habits that work well with the image models that are available now, including ChatGPT Image 2, Seedream 4.5, Flux 2 Pro, Flux.1 Kontext Pro, Recraft V4, Google Nano Banana 2, and Google Nano Banana Pro (2K).

Start clear and specific

Describe:

• the main subject
• the setting
• the visual style
• the lighting or mood
• the composition when it matters

Keep prompts concrete

Concrete wording usually works better than vague wording. Instead of "make it cool," describe what "cool" means: cinematic lighting, neon palette, wide-angle framing, minimalist illustration, and so on.

Add style only when useful

Helpful style cues include:

• photorealistic
• minimalist
• flat illustration
• cinematic
• product photography
• editorial

Too many conflicting style cues can make results worse.

Iterate in small steps

If the first result is close, change one thing at a time:

• background
• lighting
• pose
• color palette
• framing

Use editing models when the goal is revision

If you already have an image and want to change part of it, switch to an editing-capable model such as Flux.1 Kontext Pro, ChatGPT Image 2, or Google Nano Banana 2 instead of regenerating from scratch.

Let Magicdoor enhance the prompt

Magicdoor can improve prompts automatically before image generation. That is useful when you know the outcome you want but do not want to write a highly structured prompt by hand.

For model-by-model guidance, see the image model comparison.