We’re seeing more and more clients experimenting with AI image tools. Sometimes it’s a rough idea created in ChatGPT over the weekend. Sometimes it’s a surprisingly polished image made in Gemini and intended for a presentation, social post or new campaign.
The question is, can we actually use it?
Often, the answer is yes. But “use it” can mean several different things. An image might be a great way to show us what you have in mind, but it might also be unsuitable for the finished artwork. It may look perfectly fine on a phone and fall apart when enlarged. Or it might be almost there, but requires some retouching, rebuilding or preparing properly before it goes anywhere near a printer.
There is a fairly sizable gap between creating a good-looking image and producing finished marketing collateral. So, here’s our honest view of where AI imagery can be genuinely useful, where it still struggles and why there is usually a role for a designer, photographer or illustrator somewhere in the process.
What can AI image tools do now?
AI image generation has moved on considerably. Tools such as ChatGPT and Gemini can create photographs, illustrations, textures, product concepts and complete scenes from a written description.
You can specify the setting, lighting, composition and style, or ask for practical details such as space on one side for a headline. You can upload a reference image, request a different aspect ratio or ask the tool to change parts of an existing picture. Depending on the platform, it may also be possible to combine several reference images, remove objects, extend a background or place a subject in a completely different environment.
The process is conversational, so you don’t necessarily have to get everything right in the first instruction. You can ask to make the lighting warmer, move the subject further to the left or give the image a less polished feel.
That makes these tools very accessible. Someone who can explain what they are picturing can create a visual representation of it, even if they do not have the technical skills to produce the finished artwork themselves.
There are differences between the main platforms, though.
ChatGPT can generate new images and edit uploaded ones within the conversation. It can work in different styles and aspect ratios, make targeted changes and create images with transparent backgrounds. Its handling of text has also improved, although we would still avoid relying on generated wording for finished marketing artwork.
Gemini offers similar image-generation and editing capabilities. It can work from written prompts and uploaded references, combine elements and make conversational changes to a scene. It can be particularly useful when experimenting with how an existing subject might look in a different setting or creative style.
Claude is slightly different. It can analyse images, discuss visual ideas, develop briefs and help write better prompts, but it does not currently generate conventional photographs or illustrations itself. It can create diagrams, charts and code-based visuals, so it still has a place in a creative workflow, just not in quite the same way as ChatGPT or Gemini.
All these tools continue to change quickly, but at this stage, they are still giving teams a way to explore ideas easily and quickly.
Where AI imagery can work well
AI can be extremely useful at the beginning of a project, when the main aim is to decide what something should look and feel like.
It can help us explore several creative routes without committing the full budget to producing each one. We can test different compositions, settings and styles, then use those images to have a much clearer conversation with the client.
This can be useful for:
- Mood boards and early campaign concepts
- Storyboards
- Pitch documents and internal presentations
- Website mock-ups
- Blog and social media imagery
- Abstract backgrounds and textures
- Illustrative or deliberately surreal ideas
- Planning a photoshoot
- Checking whether a creative direction feels right before developing it further
We also genuinely like it when a client sends us something they have created. It does not matter whether the image is technically perfect at that point. If it helps show us what they have in their head, it has already done a useful job.
Describing an atmosphere, composition or visual style can be difficult. A rough AI-generated concept often communicates more than a paragraph of explanation ever could.
In some cases, the image may also be suitable for the final piece. That depends on its quality, content and intended use. A small digital graphic and a towering billboard are asking very different things from the same file.
Why an image that looks good on screen may not work in print
Resolution is one of the first things we check when a client supplies an AI-generated image.
Digital images are made from a fixed number of pixels. They can look sharp on a phone or laptop because they are being displayed at a relatively small size. When that same image is enlarged, those pixels have to stretch across a much larger area.
If the file does not contain enough image information, the result can appear soft, blurred or visibly pixelated.
Even so, that does not mean any small image can be placed on a billboard and expected to look good. Before we can say whether a file is suitable, we need to know:
- Its actual pixel dimensions
- The printer’s artwork requirements
- The final printed size
- How much of the layout it will cover
- Whether it needs to be cropped
- Whether the original contains compression, noise or generated artefacts
This is why we cannot judge an image simply by looking at an email attachment. We need to assess it at the size and in the format in which it will actually be used.
Changing the dpi figure in a file does not create more genuine detail. It cannot recover detail that was never captured or generated in the first place.
Sometimes an image can be enlarged and carefully retouched. Sometimes parts of it need rebuilding. In other cases, the best solution is to treat it as a visual reference and recreate the image properly.
A generated image is not the same as editable artwork
Most AI-generated images are supplied as flat raster files, such as a JPEG or PNG. They are much like photographs: all the elements have been merged into one image.
Professionally constructed artwork may contain separate layers, editable text, isolated objects and vector elements. This allows a designer to make precise changes and adapt the piece for different applications.
With a flat AI image, a seemingly simple request can be much harder than it sounds:
- Can the product be moved slightly to the left?
- Can we change only the colour of the jacket?
- Can the same person appear in another pose?
- Can we remove the wording but leave the background untouched?
- Can this square image become a very wide exhibition graphic?
- Can we use the same scene across six pieces of campaign material?
Conversational image editing is improving, but changing one part of an image can still affect something else. A face may alter slightly, a product can acquire new details or the background may be regenerated rather than simply extended.
If the campaign needs consistency across several formats, a designer may need to separate and rebuild elements so that the artwork can be controlled properly.
Keep logos and campaign copy out of the generated image
AI tools have become noticeably better at producing readable text, but “better” does not mean dependable enough for finished artwork.
You may still find a missing letter, an almost-correct logo or a typeface that changes halfway through a word. Even if it looks right at first glance, it may not match your actual brand assets.
Wherever possible, we would generate the visual without the headline, logo, call to action or small print. Those elements can then be added separately by a designer.
That gives us control over the typography, positioning, spacing, accessibility and brand consistency. It also means the wording can be changed later without having to generate the whole image again.
The same principle applies to logo design. AI can help explore broad themes or visual ideas, but it should not be relied on to produce a finished identity. A professional logo needs carefully considered typography and spacing, as well as accurate vector files that can be scaled from a small digital icon to signage without losing quality.
If an AI tool produces an interesting logo concept, a designer can use that as a discussion point and create an original, properly constructed identity from it. The generated image itself is not the finished logo.
Real products need to remain real
AI can produce an incredibly polished product image. The problem is that it may not be an accurate one.
A buckle might change shape. A label may contain invented wording. The material could look smoother, thicker or more reflective than it is in real life. On a technical product, the tool may confidently add a part that does not exist or remove one that is essential.
If the image is being used to sell a specific product, customers need to see an honest representation of what they will receive.
There is a difference between taking a genuine product photograph and using AI to help place it in a new setting, and asking AI to recreate the entire product from scratch. Even when a reference image is supplied, the result needs to be checked closely.
For ecommerce, manufacturing, food, property and other sectors where accuracy matters, professional photography will often remain the safest and strongest option. AI may still assist with concept development, background extension or small retouching jobs, but it should not quietly change what is being sold.
Brand consistency is about more than colours
An AI tool does not automatically understand the wider context in which an image will appear.
It does not know how the campaign relates to your website, brochures, social content and exhibition materials. It may not follow your usual photographic style, typography, colour values or approach to representing people. It also does not know which aspects of your identity can flex and which need to remain consistent.
You can provide reference material and detailed instructions, and the tools are getting better at following them. However, someone still needs to decide whether the result actually feels like the brand.
That is particularly important across a larger campaign. One AI-generated image can look great in isolation while sitting awkwardly beside everything else you would usually produce.
Accuracy, authenticity and trust
Not every problem with an AI image is obvious.
Generated machinery may be assembled incorrectly. PPE can be missing or unsuitable. A building might look convincing but bear no relation to a real site. Reflections, shadows and small background details can reveal inconsistencies that are easy to miss when the image is first viewed.
AI-generated people and places also need careful consideration. The imagery may unintentionally rely on stereotypes, create an unrealistic picture of an organisation or imply that a fictional situation really happened.
There are plenty of campaigns where a generated person or imagined setting is perfectly appropriate. There are others where using real employees, customers, products or locations is an important part of the story.
For businesses built around personal service, craftsmanship or a strong sense of place, genuine photography may communicate far more than a technically impressive generated image.
There may also be occasions when an image should be labelled as AI-generated or substantially altered, particularly if someone could reasonably mistake it for evidence of a real event, place, product or person.
What about copyright and commercial use?
The legal position around AI-generated imagery is still developing, and the terms vary between platforms.
Before an image is used commercially, particularly for packaging, advertising or a large paid campaign, it is worth checking how it was created and which service was used. A free account may not offer the same terms as a paid business service.
Care is also needed when prompts refer to recognisable characters, competitor branding, existing campaign artwork or the distinctive style of a living artist. The fact that a tool has produced an image does not automatically make every use of it sensible or risk-free.
For significant campaigns, we would always rather raise these questions during production than after the artwork has been published.
Where the human skills come in
Once the direction has been agreed, there is usually still a fair amount of work between an AI-generated concept and finished marketing collateral.
Depending on the project, our design team may need to:
- Retouch strange details and visual artefacts
- Rebuild missing or inaccurate elements
- Redraw graphics as scalable vector artwork
- Separate subjects from their backgrounds
- Adjust the composition for different formats
- Add brand typography and campaign messaging
- Match the organisation’s colours and visual style
- Convert and manage colour correctly for print
- Prepare artwork to the printer’s specifications
- Check that every element is accurate and appropriate
There are also occasions when we will recommend replacing the generated image altogether. A commissioned photograph, bespoke illustration or carefully constructed piece of artwork may give the business something more distinctive, more accurate and much easier to use in the long term.
That does not mean the AI concept was wasted. It may have helped establish the lighting, composition, mood or idea behind the finished piece. It has simply done a different part of the job.
So, can we use your AI-generated image?
Send it to us, and we will have a look.
Sometimes the answer will be yes. Sometimes we can improve it through careful retouching or upscaling. Sometimes we will need to rebuild parts of it, and sometimes we will suggest recreating the idea through photography, illustration or design.
For us, that is the most useful way to approach AI imagery. We are not precious about where an idea comes from. If you have created something that helps us understand what you are picturing, it gives the project a useful starting point.
Our job is to work out how to turn that starting point into something accurate, on-brand and properly made for wherever it needs to appear.
AI has made it much quicker to show people an idea. Making that idea work in the real world still takes experience, judgement and, quite often, a fair amount of human skill.