Universal prompt structure that works across all models
Every strong AI image prompt follows this core structure regardless of which model you use: [Medium] + [Subject] + [Style] + [Lighting] + [Composition] + [Quality]. Medium defines whether the output is a photograph, illustration, painting, or 3D render. Subject describes what appears in the image with specific visual detail. Style sets the aesthetic direction through art movement, photography genre, or visual reference. Lighting shapes mood and dimension through named setups or environmental light descriptions. Composition defines camera relationship to the subject through lens, angle, and framing terms. Quality sets the technical standard through resolution and detail keywords. You do not need to fill every slot for simple images, but knowing the structure ensures you never accidentally leave out critical information that could improve your results.
Quick keyword reference by category
Lighting keywords: golden hour, blue hour, Rembrandt lighting, butterfly lighting, split lighting, rim light, backlight, soft diffused, harsh directional, overcast, window light, neon glow, volumetric, chiaroscuro, three-point setup. Composition keywords: rule of thirds, leading lines, symmetrical, negative space, close-up, wide angle, overhead flat lay, eye level, low angle heroic, Dutch angle, framing within frame. Quality keywords: 8K resolution, sharp focus, highly detailed, professional photography, editorial quality, masterpiece, clean composition, film grain, shallow depth of field, bokeh. Style keywords: cinematic, editorial, photorealistic, hyperdetailed, minimalist, vintage, retro, dark moody, bright airy, warm tones, cool tones, desaturated, high contrast. Camera keywords: 85mm portrait lens, 35mm street photography, 24mm wide angle, 50mm standard, macro close-up, medium format, shot on Hasselblad, Sony A7, Leica M.
Model-specific parameter cheat sheet
Midjourney parameters: --ar 16:9 (aspect ratio), --stylize 750 (artistic interpretation 0-1000), --chaos 30 (variation randomness 0-100), --quality 2 (rendering quality), --no text watermark (negative terms), --seed 12345 (reproducibility), --tile (seamless patterns), --stop 80 (partial rendering). Stable Diffusion settings: CFG Scale 7-9 (prompt adherence), Steps 25-35 (DPM++ 2M Karras sampler), Resolution 512x768 for SD 1.5 or 1024x1024 for SDXL, Negative prompt: blurry, low quality, deformed, bad anatomy. DALL-E 3: use full descriptive sentences, specify size as square/landscape/portrait, iterate through ChatGPT conversation, ask to show the exact prompt used. Gemini: works with both keyword and sentence styles, specify aspect ratio in description, strongest at text rendering and photorealism.
Negative prompt templates ready to copy
Universal base negative prompt: blurry, low quality, worst quality, watermark, text, signature, cropped, deformed, out of frame, poorly drawn, jpeg artifacts, low resolution. For photorealistic work add: cartoon, anime, painting, illustration, CGI look, plastic skin, overprocessed. For anime and illustration work add: photorealistic, photograph, 3D render, realistic skin, photograph texture. For product photography add: tilted product, incorrect reflections, floating shadows, background artifacts, wrong proportions. For portraits add: bad anatomy, extra fingers, fused fingers, deformed iris, asymmetric face, poorly drawn hands, duplicate features. Start with the universal base and layer on category-specific terms. Keep your negative prompt shorter than your positive prompt to avoid over-constraining the generation.
Troubleshooting common prompt problems
Images too generic: add more specific style references and named lighting setups instead of vague quality keywords. Colors too saturated: add muted tones, desaturated palette, or natural color science. Hands look wrong: add natural relaxed hand pose, anatomically correct fingers, and put hand anatomy terms in your negative prompt. Text unreadable: DALL-E 3 and Gemini handle text best, specify the exact text in quotes, font style, and placement. Composition too centered: add rule of thirds positioning or off-center subject placement. Image too busy: reduce the number of described elements and add negative space, minimal composition, or clean background. Style inconsistent: choose one primary style reference and remove competing terms. Lighting flat: replace nice lighting with a specific named setup like Rembrandt side lighting with key light from upper left.