Prompt Profiles for Image AI: Midjourney, DALL-E, SD

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/nidhinjs/prompt-master/llms.txt

Use this file to discover all available pages before exploring further.

​

Midjourney

• Comma-separated descriptors, not prose — lone samurai, heavy rain not a lone samurai standing in heavy rain
• Subject always first — the model weights early tokens more heavily
• Parameters always last — --ar, --v, --style, --chaos, --q go at the very end, never inline
• Negative prompts via --no — --no text, watermark, signature, blurry (not a separate negative prompt field)
• Aspect ratio with --ar — --ar 16:9 (landscape), --ar 9:16 (portrait), --ar 1:1 (square)
• Version flag — --v 6 is the current standard; --v 6.1 for sharper details
• --style raw — reduces Midjourney’s aesthetic processing; use when you want precise descriptor adherence

​

Example

lone samurai standing in heavy rain at night, traditional armor, neon reflections on wet cobblestone street, cinematic lighting, dramatic shadows, fog, ultra detailed, photorealistic, shallow depth of field --ar 16:9 --v 6 --style raw

​

Parameter Reference

​

DALL-E 3

• Write prose, not keyword lists — A lone samurai stands in a rain-soaked alley not samurai, rain, alley
• Describe foreground, midground, and background — layered spatial descriptions produce more compositionally coherent images
• Add do not include text unless specified — DALL-E 3 will generate signs, labels, and readable text if any text-like concept is in the prompt
• Describe mood through scene details — rain-slicked cobblestones reflecting a distant neon sign beats moody as a standalone descriptor
• State the art style explicitly — photorealistic, oil painting, watercolor illustration, flat design vector

A lone samurai stands at the entrance of a rain-soaked alley at night. In the foreground, puddles on the cobblestone street reflect streaks of red and blue neon light from a distant sign. The samurai's traditional armor glistens with rain. In the background, fog obscures the alley's end, with only faint lantern light visible. Cinematic lighting, photorealistic style, dramatic shadows, shallow depth of field. Do not include any text, symbols, or writing in the image.

​

Stable Diffusion

• 20–30 steps — drafts, iteration, fast previews
• 40–50 steps — final renders, high-detail output

• (word:weight) syntax for emphasis — (ultra detailed:1.3), (bokeh:1.2)
• Negative prompt is MANDATORY — always output a negative prompt block
• CFG 7–12 — 7 for artistic freedom, 10–12 for strict prompt adherence
• Steps 20–30 drafts / 40–50 finals

(lone samurai:1.2), standing in heavy rain, traditional armor, (neon reflections:1.1) on wet cobblestone street, cinematic lighting, dramatic shadows, fog, (ultra detailed:1.3), photorealistic, (shallow depth of field:1.1), 8k resolution, masterpiece

(deformed:1.4), (blurry:1.3), bad anatomy, extra limbs, watermark, text, signature, lowres, jpeg artifacts, cropped, worst quality, low quality, normal quality, username, artist name

​

SeeDream

• Specify art style first — lead with the aesthetic: anime, cinematic, painterly, concept art, ukiyo-e, noir photography
• Mood descriptors are high-signal — melancholic, ethereal, visceral, dreamlike, harsh and overexposed
• Negative prompt recommended — not strictly required, but improves output consistency
• Subject follows style declaration — cinematic noir — lone detective in a rain-soaked city street at midnight

cinematic, noir photography style — lone detective standing under a flickering streetlight, rain-soaked city street, deep shadows, high contrast, 1940s setting, fedora and trench coat, distant neon signs reflected in puddles, atmospheric fog, desaturated with amber highlights

Negative: cartoonish, bright colors, daytime, modern clothing, text, watermark

​

ComfyUI

• Node-based workflow — prompts feed into specific nodes (typically CLIPTextEncode for positive and negative conditioning)
• Ask which checkpoint first — PromptMaster will ask which model checkpoint you’re using (e.g., SDXL, SD 1.5, Flux) before generating; syntax and optimal weights differ by checkpoint
• Always two blocks — Positive prompt block + Negative prompt block, clearly labeled
• Never merge — positive and negative content must never appear in the same block

=== POSITIVE PROMPT (CLIPTextEncode — positive conditioning) ===
[positive prompt content here]

=== NEGATIVE PROMPT (CLIPTextEncode — negative conditioning) ===
[negative prompt content here]

=== POSITIVE PROMPT (CLIPTextEncode — positive conditioning) ===
lone samurai standing in heavy rain at night, traditional Japanese armor, neon reflections on wet cobblestone street, cinematic lighting, dramatic shadows, volumetric fog, ultra detailed, photorealistic, shallow depth of field, 8k, masterpiece, best quality

=== NEGATIVE PROMPT (CLIPTextEncode — negative conditioning) ===
deformed, blurry, bad anatomy, extra limbs, watermark, text, signature, low quality, worst quality, jpeg artifacts, cropped, out of frame, duplicate

​

Reference Image Editing

​

Edit vs. Generate Detection

• Attach the reference image first — the image is the primary input; the prompt is secondary
• Prompt around the delta only — describe what should change and what should stay the same; do not re-describe the entire image
• Explicitly state what is preserved — Keep the subject's face, pose, and clothing unchanged
• State the change precisely — Replace the background with a mountain landscape at dusk

A woman with brown hair in a red dress standing in front of mountains at dusk, 
golden hour lighting, photorealistic

Reference image attached.
Keep: subject's face, pose, red dress, and lighting on the subject unchanged.
Change: Replace the background with a mountain landscape at dusk. 
Match the existing lighting direction on the subject to the new background.