Most prompt generators show you one output and hide the template. Ours assembles three strings from four parts, and every part is visible in the source. This article takes the template apart so you know exactly what you paste into Midjourney, DALL-E 3, or Stable Diffusion, including two translation failures we verified in our own dictionary code.
The four parts of every prompt
The generator joins three text blocks and appends one ratio:
1. Subject. Your idea, run through a Turkish to English word replacement first.
2. Style block. One of eight presets, appended as comma separated tags.
3. Quality block. One of three tiers, appended the same way.
4. Aspect ratio. One of six values, 1:1, 16:9, 4:3, 9:16, 21:9, or 3:4.
The defaults on load are the Realistic style, the HD tier, and a 16:9 ratio. The join is a plain comma and space, in the order subject, style, quality. Nothing smarter happens between the parts.
The three output formats, exactly
The same inputs render three different strings, and the differences are the reason to care:
Midjourney: core, then --ar 16:9 --q 2 --v 6.1
DALL-E 3: A subject, style tags, quality tags, aspect ratio 16:9
Stable Diffusion: (core:1.3), ((first quality tag:1.2))
Negative: blurry, bad anatomy, text, watermark,
signature, ugly, deformed, low quality
The Midjourney suffix pins two flags you cannot edit from the interface, quality 2 and version 6.1. The DALL-E output prepends the article A to your subject, which reads wrong for plural subjects. The Stable Diffusion output wraps the core in a weight of 1.3, weights the first quality tag at 1.2, and appends a fixed negative prompt of eight terms you cannot remove.
The tag blocks, quoted from the source
Style presets are hardcoded tag strings. The Realistic preset ships photorealistic, ultra detailed, 8k, DSLR, professional photography, sharp focus. The Cinematic preset ships cinematic, dramatic lighting, film grain, anamorphic lens, epic composition. Six more presets cover anime, fantasy, cyberpunk, oil painting, minimal, and dark art.
Quality tiers stack on top. Standard adds high quality, detailed. HD adds 8k resolution, ultra detailed, hyperrealistic, award winning. Ultra adds 8k resolution, ultra hyperrealistic, masterpiece, best quality, intricate details, perfect anatomy, professional.
Note the overlap. The Realistic style and the HD tier both carry 8k and ultra detailed, so the combined prompt states them twice. Duplicated tags are a known habit of template generators, and ours is no exception. Read the output and delete a duplicate when it bothers you.
The dictionary and the two failures we verified
Translation is a 68 entry lookup that walks the dictionary in insertion order and replaces every hit as a substring, case insensitive. Single covered words mostly work. Uçan kedi becomes flying cat. The failures come from the mechanics:
1. Substring matching hits inside longer words. Işıklar contains ışık, so the phrase neon ışıklar renders as neon lights dramatic lightinglar, with a Turkish plural suffix glued onto an English replacement.
2. Order collisions corrupt two words we tested. The ay entry, meaning moon, fires before the kaya entry, meaning rock, so kaya becomes kmoona. The same ay entry also fires on earlier output. Gündüz first becomes daytime, then ay hits inside that result, and the final string is dmoontime.
Any Turkish word outside the dictionary passes through untouched, so büyülü, meaning magical, survives into an English prompt. Write your subject in plain English and the dictionary never runs at all. That is the reliable path.
Steps to a prompt you can defend
1. Write the subject in English, or use dictionary-covered Turkish words only.
2. Pick the style preset that matches the look you want.
3. Pick the quality tier, knowing Standard adds the least and Ultra adds the most.
4. Choose the ratio that matches your target surface, 9:16 for vertical, 16:9 for wide.
5. Generate, then read all three strings before pasting anywhere.
6. Use the copy button on each card. It falls back to a hidden textarea method when the clipboard API is unavailable.
Checklist before you paste
- Subject contains no untranslated Turkish words.
- You removed duplicated tags where style and quality overlap.
- The Midjourney version flag, 6.1, matches the version you actually run.
- You fixed the A prefix on plural subjects in the DALL-E output.
- The fixed negative list fits your subject, because you cannot edit it in the tool.
Limits you should know
The tool assembles strings. It calls no model, previews no image, and validates nothing about your target engine. The negative prompt, the two weights, and the two Midjourney flags are constants in the source. The dictionary replacement is naive by design, with no stemming and no word boundaries. Treat every output as a first draft that you proofread, in the same way you proofread a generated paragraph before sending it.
Build one prompt with the
[AI prompt generator](https://webrecast.com/en/ai-prompt) and take the three outputs to
their named engines. If a model rejects a flag or a weight from the template, tell us which
engine and version, so we can pin the constants to something current.
