Image generation basics
Composition: direct the frame
Subject tells the model what is in the image; composition tells it where the viewer looks and how the image works as a frame.
Before you start
Why this matters
Move beyond lists of objects
A prompt can contain the right objects and still produce the wrong image. A person, laptop, window, and plant might all appear, yet the person may be tiny, the laptop may hide their hands, and the bright window may pull attention away from the subject. This is a composition failure.
Composition organizes visual relationships. It controls hierarchy, framing, viewpoint, depth, balance, and empty space. You do not need formal art training to direct it. You need concrete language about where the subject sits, how much of it is visible, and what should attract attention first.
Compare:
A baker decorating a cake in a bright kitchen.
with:
Medium close-up at counter height, the baker placed on the right third, hands and cake fully visible in the foreground. The face is the primary focal point and the icing bag creates a diagonal toward the cake. Softly blurred kitchen behind; uncluttered left third reserved for title copy.
The second prompt gives the model a spatial plan.
1Learn the idea
Read
Choose the shot size
Shot size determines how much context surrounds a subject:
- Extreme close-up: a small detail fills the frame, such as a brush touching paper.
- Close-up: a face or object dominates; useful for expression and texture.
- Medium shot: often frames a person from waist or chest up; balances action and identity.
- Full-body shot: shows posture, clothing, and interaction with nearby space.
- Wide or establishing shot: emphasizes environment, scale, and location.
- Overhead or top-down: clarifies layouts, ingredients, desks, and arrangements.
Use a shot size because it serves the message. A close-up communicates emotion but may hide a workplace process. A wide shot provides context but can weaken facial detail. If a particular body part or object must remain visible, say so directly: “both hands fully in frame” or “entire bicycle visible, including both wheels.”
Read
Set viewpoint and lens character
Viewpoint changes power and intimacy. Eye level often feels neutral. A low angle can make a subject imposing. A high angle can reveal a layout or make a scene feel observed. An over-the-shoulder view connects a person with what they see.
Lens terms can communicate visual character, but use them intentionally:
- A wide-angle view exaggerates depth and includes more environment.
- A normal perspective feels relatively natural.
- A telephoto look compresses distance and isolates subjects.
- Shallow depth of field separates a subject with background blur.
- Deep focus keeps foreground and background readable.
Exact focal lengths may not behave like a physical camera in every model. Treat them as directional cues, then judge the visible result. “24 mm” is less important than whether the frame actually feels expansive and whether edge distortion is acceptable.
Read
Direct visual hierarchy
Hierarchy answers: what should the viewer notice first, second, and third? You can create it through scale, contrast, light, color, sharpness, and placement.
State the primary focal point and reduce competition around it:
The yellow rescue kayak is the brightest element and the clear focal point. Dark blue water surrounds it; the distant shoreline is low contrast and secondary.
For layouts, name zones. “Product centered” is useful; “product centered in the lower half with the upper 40 percent simple and low contrast for white copy” is more operational. If a design must crop across devices, protect a safe central region and keep expendable details near the edges.
Common organizing patterns include:
- Rule of thirds: subject offset for movement or copy space.
- Centered symmetry: stable, iconic, or product-focused.
- Leading lines: roads, shelves, shadows, or gestures guide attention.
- Frame within a frame: doorways or foliage create depth.
- Layered depth: foreground, middle ground, and background establish scale.
- Flat lay or grid: useful for comparison and catalog-like clarity.
These are tools, not rules. Name the effect you need even if you do not know the formal label.
Read
Compose for downstream use
An image is often one layer in a larger design. A website hero needs responsive crops. A social post may need space for a logo added later. A slide background must not compete with text.
Include layout constraints such as:
- “No important details in the outer 10 percent.”
- “Simple dark area on the left for high-contrast white text.”
- “Subject remains readable when cropped to square.”
- “Background low-detail enough for presentation copy.”
- “No generated text; typography will be added separately.”
Generate at the closest supported aspect ratio, but plan for final cropping. Extending a canvas or using outpainting can help, yet newly generated edges may introduce inconsistent objects. Recheck the full image after every expansion.
Read
Diagnose, then revise
Do not respond to a weak composition with “make it better.” Identify the spatial failure:
- Subject too small → request a closer shot or larger subject scale.
- Frame too busy → simplify the background and reduce secondary objects.
- Wrong emphasis → specify focal point, contrast, and lighting.
- No copy space → assign the subject to a third and reserve a quiet zone.
- Awkward crop → name what must be fully visible and add safe margins.
- Flat scene → add foreground, middle ground, background, or directional light.
Change one or two composition variables at a time. If you simultaneously change angle, palette, subject, and medium, you will not know which instruction improved the frame.