AI Prompt Architect preview

AI Prompt Architect

Build image prompts for Midjourney, DALL-E 3 and Stable Diffusion. Describe your idea, pick one of 8 styles, a quality level and an aspect ratio, and copy a ready prompt for each model.

Key features

  • 8 style presets, from Realistic and Cinematic to Oil Painting and Dark Art
  • Separate prompts for Midjourney, DALL-E 3 and Stable Diffusion
  • Aspect ratios 1:1, 16:9, 4:3, 9:16, 21:9 and 3:4, with a copy button per prompt
  • Stable Diffusion output includes weights and a negative prompt

Guide

An image generator turns a short text description into a picture, and the words you choose decide most of the result. Midjourney, DALL-E 3 and Stable Diffusion all read prompts differently. Midjourney expects comma-separated keywords followed by parameters such as --ar for the aspect ratio. DALL-E 3 works best with a plain sentence that describes the scene. Stable Diffusion accepts weights in brackets and a separate negative prompt that lists what to leave out. The WebRecast AI Prompt Architect takes one idea from you and writes a version for each of these three models.

How the tool works

You type your idea into the text box, for example "Gothic castle, fog, red moon". You can also click one of the four example ideas under the box to fill it in. Then you choose three settings. The style decides the look of the image. The quality level adds detail keywords. The aspect ratio sets the shape of the picture. When you press Generate Prompt, three cards appear: Midjourney, DALL-E 3 and Stable Diffusion. Each card has its own Copy button, so you can paste the prompt straight into the model you use.

The tool runs in your browser. It does not send your idea to any AI service and it does not generate images. It only builds the text that you paste into an image generator.

The 8 style presets

Each style adds a fixed set of keywords to your idea. Realistic adds photorealistic, ultra detailed, 8k, DSLR, professional photography and sharp focus. Cinematic adds dramatic lighting, film grain, an anamorphic lens look and epic composition. Anime adds manga and illustration keywords with vibrant colors. Fantasy adds magical, ethereal and glowing keywords with intricate detail. Cyberpunk adds neon lights, a futuristic city and a dark atmosphere. Oil Painting adds impressionist brushstrokes and a classical look. Minimal adds clean design, negative space and flat illustration. Dark Art adds gothic, moody and chiaroscuro keywords.

Pick the style that matches the final use. A product mockup or a stock-style photo fits Realistic. A poster or a book cover often fits Cinematic or Fantasy. An icon or a simple header graphic fits Minimal.

The 3 quality levels

Standard adds high quality and detailed. HD adds 8k resolution, ultra detailed, hyperrealistic and award winning. Ultra adds a longer list that includes masterpiece, best quality, intricate details and perfect anatomy. HD is selected when the page opens. These words do not change the pixel size of the output. They push the model toward a more detailed rendering style. If your images look overworked or too busy, switch to Standard.

The 6 aspect ratios

You can choose 1:1, 16:9, 4:3, 9:16, 21:9 or 3:4. The tool selects 16:9 when the page opens. Use 1:1 for profile pictures and square social posts. Use 16:9 for video thumbnails, desktop wallpapers and slides. Use 9:16 for phone wallpapers, stories and short video covers. Use 21:9 for wide banners. Use 3:4 or 4:3 for portrait or landscape prints.

What each output contains

The Midjourney prompt is your idea, the style keywords and the quality keywords, joined by commas, followed by --ar with your ratio, --q 2 and --v 6.1. If you use a different Midjourney version, edit the --v value after you paste it.

The DALL-E 3 prompt starts with "A", then your idea, the style keywords, the quality keywords and the phrase "aspect ratio" with your choice. DALL-E 3 reads the ratio as text, so check the size setting in the app you use as well.

The Stable Diffusion prompt wraps the full keyword list in brackets with a weight of 1.3 and adds the first quality keyword with a weight of 1.2. A second line starts with "Negative:" and lists blurry, bad anatomy, text, watermark, signature, ugly, deformed and low quality. Paste the first line into the prompt field and the words after "Negative:" into the negative prompt field of your interface.

Turkish input

The tool has a built-in list of 68 common Turkish words for subjects, places, weather, colors and moods. If you type an idea in Turkish, words on that list are replaced with English, so "ejderha" becomes dragon and "orman" becomes forest. Words that are not on the list stay as you typed them. Only whole words are replaced, so an English word like "day" stays as it is. A Turkish word with a suffix, such as "kediler", is not on the list and stays as you typed it. Read each prompt before you copy it.

Writing a better idea

The tool adds style and quality keywords for you, so your input only needs to describe the content of the picture. Start with the main subject. Add what it is doing, where it is, and the time of day or weather. "Flying cat, neon lights, cyber city" gives the model a subject, a light source and a setting. "A cat" gives it almost nothing to work with.

Keep your idea short. Five to ten keywords or one sentence is enough. Long inputs can bury the subject among the style keywords, and the model may lose track of what matters.

Avoid mixing styles in the idea itself. If you type "anime" in the idea and pick the Realistic preset, the prompt contains both photorealistic and anime keywords, and the result can look confused. Let the preset handle the style.

Testing and refining

Generate a few images with the first prompt and look at what went wrong. If the subject is too small, move it to the start of your idea. If the image is too dark, add a light source such as sunset or neon lights. If unwanted text appears in Stable Diffusion results, the negative prompt already lists text and watermark, so increase the negative weight in your interface or rephrase the idea.

Try the same idea with two or three styles. The keywords change a lot between presets, and comparing results is the fastest way to learn what each model does with them. You can change any setting and press Generate Prompt again, and the three cards update with the new version.

Edit the output when you need to. The prompts are plain text, so you can add a camera angle such as close-up or wide shot, or a color palette before you paste them into the model.

Frequently asked questions

What makes a prompt "good"?

A clear subject, a style, and details such as lighting, weather or setting. The tool adds style and quality keywords for you, so focus your input on the subject and the scene.

Which AI models does this work with?

The tool formats prompts for Midjourney, DALL-E 3 and Stable Diffusion. It builds image prompts only, so it is not designed for text chatbots.

Related guides

Related WebRecast sections