Lesson 2. Visual prompt writing technique: subject, style, lighting, and composition
When an AI image disappoints, many think the model is to blame. In reality, the problem in most cases is the prompt. A visual prompt consists of five key elements — the right combination improves quality 2-3 times.
Topic breakdown
A prompt for text AI (ChatGPT) and for visual AI are different things. For text, role, context, and format are important. For visual — subject, aesthetics, lighting, and composition.
A good visual prompt is a brief for an artist. You describe every important element of the image, and the model translates the description into pixels.
The five key elements: subject (what), art style (what look), lighting (what light), color (what mood), composition (what angle).
We will study each element separately, then practice combining them into one prompt.
What you'll learn
- Know the 5 key elements of a visual prompt
- Master the most useful keywords for each element
- Write adapted prompts for different purposes
- Consistently improve the prompt through iteration
Deep dive
Think of a visual prompt not as one long sentence, but as a set of layers: first the subject, then aesthetics, then camera or light, and finally color and composition. This approach makes editing faster and cleaner.
In a good prompt, almost every word works. If an adjective doesn't change the result, remove it and replace it with a more concrete signal that actually helps the model.
If the picture turns out chaotic, don't rush to simply lengthen the prompt. First, fix the composition: 'centered product shot', 'close-up portrait', or 'wide cinematic frame' often provide the most noticeable shift.
Lesson plan
Subject: the foundation of the image
Without a concrete subject, the model guesses. Instead of 'A dog', write 'a golden retriever puppy sitting on a porch step, looking at the camera'.
Art style: visual language
Without a style, the model chooses a standard one — usually digital art or photography. 'Pencil sketch', 'Baroque painting', 'isometric illustration' give strong results.
Lighting: atmosphere creator
Lighting is the secret of professional photos. 'Golden hour', 'neon rim light', 'overcast diffused', 'candlelight' — each creates a different feeling.
Color and composition together
Color palette and composition work in pairs. 'Muted pastel palette, centered composition, soft focus background' — these three words completely change the image.
Weak vs strong prompt
a coffee shop
cozy coffee shop interior, warm bokeh lights, wooden tables, steam rising from cups, morning light through large windows, photographic style, shallow depth of field, warm amber tones, wide angle shot
The second prompt specifies over ten specific elements: atmosphere, lighting, color, composition, style. The first gives the model complete freedom — the result is random.
Ready visual prompt template
Copy and adapt[subject and details], [art style], [lighting type], [color palette], [composition and angle], [quality keywords]
Why it works
Subject — who or what is depicted. Write concretely: instead of 'a woman' — 'a young woman in her 20s wearing a red dress'.
Art style gives the visual language: 'oil painting', 'photography', 'digital art', 'watercolor', 'concept art'.
Lighting determines 60% of the mood: 'golden hour', 'dramatic rim light', 'soft box studio', 'foggy morning'.
Color palette: 'warm earth tones', 'cool blue palette', 'high contrast black and white', 'pastel colors'.
Composition and angle: 'wide shot', 'close-up portrait', 'bird's eye view', 'rule of thirds', 'bokeh background'.
Practice
- Take a weak prompt: 'a market in the city'.
- Add a subject: 'a bustling outdoor spice market in a Middle Eastern city'.
- Add lighting: '...golden hour sunlight filtering through awnings'.
- Add style and color: '...cinematic photography, warm amber tones, shallow depth of field'.
- Generate both versions and compare the results.
Checklist
Common mistakes
- describing only the subject, ignoring style and lighting
- overloading the prompt with adjectives, losing the main point ('beautiful amazing gorgeous stunning')
- mixing contradictory styles: 'oil painting, photorealistic, anime style'
- not specifying composition — the angle becomes random
Lesson FAQ
How many words should a prompt contain?
For Midjourney, 20-60 words is optimal. For DALL·E, 1-3 detailed sentences work well. Overly long prompts often 'blur' the most important elements.
Do visual prompts work in other languages?
They work, but English prompts yield noticeably better results. We recommend writing key visual terms in English.