Image generation tools: choosing and using them well

0:00 AI image tools in 2026

There are now dozens of AI image tools, and each one is the best at something. The skill is not knowing every tool. It is knowing which kind of tool fits your brief, how to prompt any of them well, and how to stay honest and legal with the results. That is what this lecture covers.

0:24 Why this matters

Why does this matter? Visuals drive attention and sales, and AI can cut concept rounds from days to hours. That is a huge advantage for small teams and agencies. But a visual that misrepresents a real product, or uses someone's likeness without consent, costs trust and can break advertising and consumer protection rules. So speed and responsibility have to travel together.

0:51 Tool families

Think in families. Assistant integrated generators, like ChatGPT Images and Gemini's image models, often called Nano Banana, are great at following instructions, rendering text and editing through conversation. Style focused tools, like Midjourney, excel at distinctive aesthetics and concept art. Design suite tools, like Adobe Firefly and Canva's AI features, sit inside editing and brand workflows. Specialists like Ideogram focus on legible typography. And open weight models, like FLUX variants and Stable Diffusion families, give you local control and fine tuning.

1:26 The analogy

Here is an analogy. Image tools are like cameras and lenses. A good photographer does not argue about which camera is best. They pick the right one for each shot. And the real skill, composition, light and storytelling, travels with them from camera to camera. Prompting is that skill. Learn it once, and it works in every tool.

1:51 The universal prompt

Here is a universal structure. Subject, a matte black insulated water bottle with a bamboo lid. Setting, on a rock beside a mountain stream in morning mist. Style, outdoor lifestyle product photography. Composition, bottle in the right third, space on the left for text, eye level angle. Lighting and mood, soft, calm and fresh. Format, four by five vertical. And what to avoid, logos, text and people. This works, with small adaptations, in almost every tool.

2:24 Simple example

A simple example. You want a birthday card illustration of a cat wearing a party hat. Version one is too realistic, so you change only the style to flat cartoon. Version two is great, but the colours clash with your envelope, so you change only the palette to pastel. Version three, you move the cat left to leave room for a message. Three single changes, logged, and you know exactly which prompt produced the final card.

2:57 Iterate like a designer

For professional work, iterate like a designer. Change one variable at a time. Keep a prompt log with the prompt, tool, settings, result and notes. Use reference images for consistent characters, products or styles where the tool supports it. And build a brand style sheet, approved colours, lighting, angles and example outputs, so anyone on the team can reproduce the look.

3:24 Real products need real photos

Here is the rule that matters most commercially. AI tools subtly change product shapes, labels and colours. For ads and product pages, composite real product photography into AI generated backgrounds, or use AI for concept boards and then shoot the real thing. Advertising rules in the UK, the US and across the Gulf prohibit misleading representations of products, and customers notice when the box that arrives does not match the ad.

3:55 Provenance, labels, rights

Now provenance and rights. Content Credentials, based on the C two P A standard, attach information about how an image was created or edited, and Google uses SynthID watermarking. Social platforms have rules requiring AI labels on realistic synthetic content. The EU AI Act's transparency obligations, applying from August twenty twenty six, require disclosure of certain AI generated or manipulated content, such as deepfakes. And avoid imitating living artists' distinctive styles commercially, reproducing trademarks, or depicting real people without consent. Check each tool's terms on commercial use.

4:33 Business example: UAE skincare campaign

A realistic example. A UAE skincare brand creates mood boards in Midjourney, tests layouts with headline text using a chat based generator, and produces the final ad visuals by compositing studio photos of the real products into AI backgrounds in Photoshop with Firefly. Every final asset keeps its Content Credentials, the team logs prompts and approvals, and nothing depicts a real person who has not signed a release. Fast, beautiful and defensible. They kept a prompt log and a style reference sheet for the campaign. When the brand extended the campaign into Ramadan two months later, a different designer reproduced the look on the first attempt using the log and the sheet. Consistency came from documentation, not from any single tool.

5:26 Common mistakes

Four common mistakes. Using AI renders of the product itself in ads. Changing everything at once and losing what worked. Ignoring platform AI labelling rules. And assuming every tool grants the same commercial rights. Each tool's terms are different, so check before a campaign, not after.

5:46 Pick by brief

Here is a quick way to choose. If the image needs headline text, start with an assistant integrated generator or a text specialist. For moody concept art, a style focused tool. To edit a real product photo, a design suite with generative fill. For many consistent images of the same product or character, a tool with reference images or a fine tuned open weight model. And for a confidential, unreleased product, a local or enterprise tool with clear data terms.

6:21 Watch me do it, part 1

Let me create the water bottle visual. I paste the universal prompt. Subject, a matte black insulated bottle with a bamboo lid. Setting, a rock beside a mountain stream in morning mist. Style, outdoor lifestyle photography. Composition, bottle in the right third, space on the left for text. Lighting, soft and calm. Format, four by five. Avoid, logos, text and people. Version one is good, but a little warm. Version two changes only the light, cooler and fresher. Version three changes only the composition, bottle slightly lower. I log each change as I go.

7:02 Watch me do it, part 2

Now the professional step. The AI version of the bottle has a slightly wrong lid shape, so it will not appear in the ad. Instead, I take version three's background, place a studio photo of the real bottle into it, and add a soft shadow to match the light. Then the review checklist. Product accuracy, correct, it is the real photo. Text, none. Artefacts, the stream looks natural. Brand fit, yes. Rights, no trademarks or people. Disclosure, the campaign follows the platform's AI labelling rules for edited imagery. I export with Content Credentials kept.

7:43 Recap and try this now

Recap. Choose the tool family that fits your brief. Use the universal prompt structure, iterate one variable at a time with a log, composite real products, and respect provenance, labels and rights. Try this now. Create three versions of one brand image, changing one variable each time, then run the review checklist on your favourite, product accuracy, spelling, hands and faces, brand fit, rights and disclosure.