Image Generation Foundations

Image generators can turn a sentence into a picture in seconds. That makes it easy to make a lot of mediocre images, and hard to make the specific, usable one a client needs. The difference is direction. You need to know what you want to see, describe it in terms these models respond to, and refine it on purpose. This lesson builds that foundation for the whole path.

What the models are good and bad at

• Strong: mood, lighting, texture, style, composition, concept art, backgrounds, product-style shots, illustration and fast exploration of many directions.
• Getting better, but still check: readable text inside images, hands and bodies, counting objects, keeping details the same across images, correct perspective and brand-accurate logos.
• Weak or risky: an exact copy of a real product, a real person's face (and you should not fake it), precise technical diagrams, and anything where almost right is wrong.

The anatomy of a prompt

A reliable image prompt answers the same questions in about the same order. Models give more weight to early words. So start with the subject and the most important qualities.

• Subject: what it is, with details. A weathered ceramic teapot with a chipped spout, not a teapot.
• Action or pose: what is happening.
• Setting: where it is, with context and props.
• Style or medium: photograph, watercolor, flat vector illustration, 3D render or risograph print.
• Composition and camera: wide, close-up, top-down, rule of thirds, shallow depth of field, or the feel of a lens.
• Lighting: soft window light, golden hour, hard studio flash or neon rim light.
• Color and mood: a palette, warm or cool, calm or dramatic.
• Format and exclusions: the aspect ratio, space for text, and what to avoid.

[SUBJECT with specifics], [ACTION], in [SETTING], [STYLE/MEDIUM], [COMPOSITION/CAMERA],
[LIGHTING], [COLOR PALETTE], [MOOD], [FORMAT: aspect ratio, space for copy]. Avoid: [unwanted things].

Example:
A weathered ceramic teapot with a chipped spout steaming on a rustic wooden table, morning window
light, cozy editorial food photograph, 50mm lens, shallow depth of field, close three-quarter
view, warm cream and deep teal palette, calm and nostalgic, 4:5 vertical with empty space at the top
for a headline. Avoid: text, logos, hands, plastic look.

Controls you will meet in most tools

• Aspect ratio: decide it first. If you design for 16:9 and then crop to 1:1, you lose the composition.
• Seed: a number that makes a result repeatable. The same prompt and seed give the same starting point.
• Reference images: show the model a style, a composition or a subject to follow, with a strength control.
• Negative prompts, or exclusions: some tools support them, and in others you write them into the prompt.
• Variations, inpainting and outpainting: change part of an image, or extend it, without starting again.
• Upscaling: raise the resolution for print or large displays, after you have chosen the final image.

A professional workflow

1. BRIEF      Purpose, audience, where it will be used, size/format, must-haves, must-avoids.
2. EXPLORE    Generate many cheap variations in 3-4 different directions (low-res is fine).
3. SELECT     Choose 1-3 with a client or against the brief - don't polish everything.
4. REFINE     Vary, inpaint problem areas (hands, edges, text), fix composition, match palette.
5. FINISH     Upscale, color-correct, remove artifacts, add real text/logos in a layout tool.
6. QA         Zoom to 100%: artifacts, warped details, odd shadows, wrong text, copyright traps.
7. DELIVER    Correct sizes, formats, color profile, file names; keep prompts and seeds on record.