Why does the output change by model?

Each model reads prompts differently. Midjourney uses --ar and --no flags and weights commas; Stable Diffusion uses a separate negative prompt; DALL·E prefers a single natural-language sentence. The builder formats your inputs to each model's conventions.

Where does the negative prompt go?

For Stable Diffusion and SDXL it is emitted as a separate "Negative prompt:" line, which is how those models consume it. For Midjourney it is converted to --no flags. DALL·E ignores negative prompts, so it is folded into descriptive phrasing instead.

How specific should the subject be?

Very. "A red fox" is weak; "a red fox curled asleep on moss, frost on its fur" gives the model something concrete to render. Specificity in the subject and style fields does more than any other setting.

Is anything uploaded?

No. The builder runs locally and only outputs text. Nothing is sent to any image model or stored.

Image Generation Prompt Builder

Image generation prompt builder

A great diffusion image starts with a structured prompt, and the structure differs by model. Midjourney rewards comma-weighted phrases and flags; Stable Diffusion wants a separate negative prompt; DALL·E reads a single clear sentence. This builder lets you fill in the parts of a strong prompt — subject, style, lighting, composition, camera — and then formats the result correctly for the model you are targeting.

How it works

You describe the subject (the most important field), choose a style, and layer in lighting, composition, and optional camera/lens specs for photographic realism. A negative prompt field lets you list what to avoid. When you pick a target model, the builder assembles everything in that model’s syntax: Midjourney gets --ar and --no flags, Stable Diffusion gets a separate negative prompt line, and DALL·E gets a natural-language sentence with the avoidances phrased positively.

Tips and notes

Lead with the subject. All three models weight early tokens — put the concrete subject first and the stylistic modifiers after.
Camera specs add realism. “35mm, f/1.8, golden hour” pushes the output toward photography; drop them for illustration and painting styles.
Keep the negative prompt focused. For Stable Diffusion, a tight list (“blurry, extra fingers, watermark”) beats a wall of tokens.
Iterate one field at a time. Change only the lighting or only the style between runs so you can tell what each modifier actually does.