AI for graphic designers: a practical, honest guide for 2026

AI for graphic designers is not one tool and not one answer. It is a set of fast, uneven tools that are very good at producing options and at the tedious parts of production. They are still weak at what clients actually pay a designer for: a clear idea, correct Arabic letters, a brand that looks like itself everywhere, and a sense of what this client needs in this market.

So the honest answer to "should designers use AI?" is yes, as a tool you direct. The job moves towards direction and judgement. You write the brief, choose from the options, fix what the tool gets wrong, and own the result.

I have been designing since 2010, and a large part of my work is brand identity, Arabic lettering and packaging. Today I direct AI tools for image, video and voice in my video work at SkillUp MENA, and I build products with Claude Code. This guide covers what AI does well and badly at each stage of a design job, where your judgement stays essential, a realistic workflow for a brand or packaging project, the licensing and copyright points worth checking, and how to start.

What AI does well and badly: the short version

The sections below go through each row of this table.

Ideation: more angles, not better ideas

A chat assistant such as ChatGPT, Claude or Gemini is a good sparring partner at the start of a project. Ask it for metaphors for a brand name, for the visual clichés of a sector, or for the questions a brief forgot to answer. You get breadth in minutes.

The limit is built into how these models work. They tend to offer the most likely answer first, and in branding the most likely answer is the one your client's competitors already have. Coffee becomes a bean and a cup. Real estate becomes a roof. Technology becomes a circuit board.

So read the list as a map of clichés: it shows you what to avoid. The idea itself should come from the brand: its name, its product, its origin or its promise. Write it in one sentence before you generate a single image. If you cannot, no prompt will find it for you.

Logos are where people expect the most from AI, so I wrote a separate guide on designing a logo with AI and where logo generators fail.

Moodboards and visual direction

Moodboards are one of the most useful jobs for AI images. Stock libraries rarely have exactly the mood you mean: a roastery in an old Cairo street at dawn, or a majlis lit by a single lantern. A generated image can show that mood in minutes, so you and the client agree on a direction before anyone draws.

Some tools are built for this. Midjourney, for example, has style references and personal moodboards that steer new images towards a set of pictures you choose, according to its documentation.

The risk is that a generated moodboard looks finished, and a client can fall for one picture and expect the design to match it. Three rules help:

• Mix generated images with real references: photography, materials, type specimens.
• Label the board as direction, not design.
• Say which parts will really be designed or photographed.

Image generation: useful, but not for everything

Each image generator is known for something different, according to the company that makes it:

• Midjourney is a dedicated image generator with its own style controls.
• ChatGPT Images creates and edits images inside a conversation, including adding text and making a background transparent, according to OpenAI's help pages.
• Google's Gemini image models, widely known as Nano Banana, focus on editing an existing image while keeping people and objects consistent.
• Adobe Firefly is trained, Adobe says, on licensed content such as Adobe Stock and on public-domain material, and Adobe attaches Content Credentials to images fully generated with it.

They are good at concept scenes, backgrounds, textures and storyboard frames. They struggle with exact product details (the real shape of a client's bottle, the real label), with keeping an object identical across many images, and with text. I compare them by criteria, not hype, in AI image generators for designers.

The skill that makes the biggest difference is not the tool. It is the prompt, written like an art director's brief: subject, framing, light, material, mood, and what to leave out. That deserves its own article: prompt writing for designers.

Retouching, generative fill and expand

This is where AI saves time quietly. Photoshop's Generative Fill adds or removes content in a selected area from a short text prompt, and Generative Expand fills new canvas when you crop outwards, according to Adobe's help pages. Removing a cable from a shop photo or extending a background so a horizontal photo fits a vertical story used to need careful cloning. Now it goes much faster.

Two cautions. The fill invents: on product shots it can quietly change text, a logo or a label detail, so check every generated area at full size. And these features need an internet connection and use generative credits, so check your plan before promising a fast turnaround.

Mockups and scenes around your real design

Mockups suit AI well, as long as the design itself stays real. The tool creates the scene: a café counter, a hand holding a pouch, a shelf in soft light. Your artwork goes onto the product as a real layer that you place, not something the tool redraws.

On the Kababgy Al-Sultan identity, some of the takeaway and menu scenes were generated with AI around the real packaging designs, and the project page says so. That is the line I would keep: the box, the bag and the menu are designed, the setting around them is generated, and the viewer is told.

Ask a tool to "put my logo on this cup" and it may redraw the logo with small errors: a curve that is not yours, a missing dot. Generate the scene, then place your artwork yourself.

Upscaling: rescue, not magic

AI upscalers enlarge an image and add detail as they go. Photoshop now has Generative Upscale, Camera Raw has Super Resolution, and standalone apps do the same job. They help with an old product photo or a generated image that needs to reach print size.

Remember what the tool is doing: guessing. The new detail was never in the original, so faces, text and fine patterns can change. And if a client's logo only exists as a tiny JPEG, upscaling is the wrong answer. Redraw it in vector.

Video and motion

Video is where AI shows most clearly in my own work. On the AI video series I direct at SkillUp MENA, the chain runs from script to stills, then motion, voice and edit, with AI as my tool at each step. Tools such as Kling and Seedance turn a still image or a written description into a short moving shot, and both makers describe features for keeping references consistent from shot to shot.

With AI as my tool, I also produced a short film for The Coffee Address, a coffee brand in Saudi Arabia. Shots that once needed a location and a crew can now start from a still and a clear direction.

What AI does not do is make the film. Keeping the same cup, hands and light between shots still takes work, text on screen belongs in the edit, and the story and the cut remain the director's job. A generated shot is raw footage, not a film.

Voice: drafts and finals, with direction

AI voice tools such as ElevenLabs read a script aloud in many languages, Arabic included, according to the company's help pages. In video work you hear the timing before anything is recorded, and on some projects the generated voice is the final one.

It still needs direction. Check how names and numbers are pronounced, and choose the register: Modern Standard Arabic for a launch film, a Gulf or Egyptian dialect for a social clip. Arabic usually leaves short vowels unwritten, so the same letters can be read as different words. Add vowel marks wherever that could happen. And never clone a real person's voice without written permission.

Copy and naming

Chat assistants are good at structure and options: an outline for a brand guide, directions for a tagline, a summary of a long brief. They are weak at a distinct voice, and they state wrong facts with confidence.

Arabic needs extra care. Generated Arabic copy can read like a translation, with phrasing no Arab copywriter would choose. Use the draft for ideas and order, then write the final Arabic yourself or with a copywriter. On packaging, the legal text (ingredients, warnings, registration numbers) must come from the client and the regulator, never from a model.

Coding small tools

This is the newest skill on the list, and it may turn out to be the most useful. AI coding assistants let a designer build small tools without becoming a developer: a script that exports every size a campaign needs, a simple landing page, a template that turns a spreadsheet into videos. I build products with Claude Code, Anthropic's coding tool that reads a project, edits files and runs commands, as its documentation describes it.

The 36 Success Stories for SkillUp MENA show the idea at a smaller scale: thirty-six videos produced from one motion template built in code. One design, many outputs, and every change made once.

The limit is responsibility. You ship what you build, so read what was written, test it, and never paste passwords or client data into a tool you do not control.

Where the designer's judgement stays essential

AI changes how fast you can make things. It does not change who is responsible for them. Five areas still depend on you.

The concept

A brand idea comes from understanding a business: who it serves, what it promises, what makes it different from the shop next door. A model can list metaphors. It cannot know which one is true for this client, and the concept should be yours.

Arabic type

This is the clearest weakness for anyone designing for Arab markets. Arabic letters join, change shape depending on their position in a word, and carry dots that change their meaning. Image models learn letters as shapes, so Arabic text in generated images often comes out disconnected, mirrored, missing dots or simply invented. Some newer models are promoted as better at text in many languages, but "better" is not the same as "correct".

For anything a client will print, set the Arabic yourself in a design app, or draw it. Calligraphic marks are even less suited: a model does not know that the word has to be read. I go through what works and what does not in Arabic text in AI images.

Brand consistency

A brand is a system: the same colours, type, spacing and tone, for years, on every surface. Generation is variation by nature. Ask twice and you get two different answers. Use AI inside the frame of the system: give it reference images, keep the real logo and type as layers you place yourself, and check every output against the guidelines.

Client context

A model does not know the client's shelf, competitors, printer, budget or the rules of their sector. Medicine packaging has regulations, and a Ramadan campaign carries meaning that a generic stock look misses. That context comes from talking to the client, and it decides more than any tool.

Ethics

A few lines worth holding to:

• Be open. Tell clients which assets are generated and with which tool.
• Label generated scenes in portfolios and presentations, the way the Kababgy page does.
• Leave living artists out of your prompts. Asking for the style of a named artist borrows their work without permission. Adobe Stock, for one, does not accept generated content with artists' names in its prompts or titles.
• Get consent for real faces and voices.
• Protect client secrets. On some services, what you generate is public by default. Midjourney, for example, offers Stealth mode for private work only on its Pro and Mega plans, according to its documentation. An unreleased product does not belong in a public gallery.

A realistic workflow: a packaging project with AI as my tool

Here is how I would run a coffee packaging project today, from brief to launch. Notice where the AI work sits: around the design, not in place of it.

Stages 2 and 4 stay human on purpose. The idea and the mark are what the client lives with for years, and they are where AI is weakest, especially with Arabic lettering.

Stage 5 is where AI earns its place. Once a sketch exists, many colour and pattern variations around it are fast and useful. You are no longer asking the tool for an idea. You are asking it to stress-test yours.

Stage 6 is not negotiable. Print needs clean vector artwork, exact colours and correct text. A generated image is a picture of packaging. It is not packaging.

Stage 10 protects everyone. A note of which assets were generated, and with which tool, answers later questions about rights.

Licensing and copyright: what to check

This is not legal advice, and the law is still moving. But a few points are clear enough to act on.

Who owns the output?

Start with the terms of the tool and of the plan you are on. They differ:

• OpenAI's terms say you own the output, to the extent the law allows, and also warn that output may not be unique: other users may receive similar results.
• Midjourney's terms give you ownership of what you create, but for larger companies they tie that ownership to its higher plans.
• Adobe presents Firefly as commercially safe, because of what it says the model was trained on: licensed content and public-domain material.

Can it be protected by copyright?

Ownership under a tool's terms is not the same as protection under the law. Here is where things stand:

• United States. In its January 2025 report on copyrightability, the US Copyright Office concluded that AI output is protected only where a human author has determined enough of its expressive elements. Prompts alone are not enough. A person's selection, arrangement and changes can be protected. In Thaler v. Perlmutter, a federal appeals court confirmed in March 2025 that copyright needs a human author, and in 2026 the Supreme Court declined to hear the case.
• Saudi Arabia. A new Copyright Law (Royal Decree M/169) and its implementing regulations, published in the official gazette Umm Al-Qura on 31 July 2026, define the author as the person who creates the work and allow copying lawfully published works to develop AI products, under conditions. I found no rule in them on who owns an image that a tool generates. Reports in 2025 describe guidance from the Saudi Authority for Intellectual Property (SAIP) that would deny protection to fully autonomous AI output and allow it for AI-assisted work with enough human contribution. Check SAIP's own publications or a lawyer before you rely on it.

The rules differ by country and are still changing, and court cases about how models were trained are still going on.

What this means in practice

• Do not promise a client exclusive rights to a purely generated image.
• Keep the parts that matter most (the logo, the lettering, the key illustration) drawn by a person.
• Keep records: sketches, prompts, versions and edits. They show the human contribution if it is ever questioned.
• For logos, trademark is a separate question from copyright.
• For anything high-stakes, the client should ask a lawyer in their own country.

How to start without losing a month

1. Pick one stage where you lose time, such as clean-up, mockups or resizing, and learn one tool for it properly.
2. Learn to brief the tool. A prompt is a brief, so the better you brief people, the faster this goes.
3. Set your lines before a client asks. Decide what you will never hand to a model. I would start with the idea, the mark and the Arabic type.
4. Keep notes on what worked, and on what failed.
5. Try one small coding project, such as a script that renames and exports files.
6. Read the terms of every tool you pay for, especially the parts on ownership and privacy.

Will AI replace designers?

Short answer: AI is taking over some tasks, such as quick retouching and simple stock-style images. It is not taking over the decisions: what the brand stands for, how it should look, and whether the work is right. I give the longer answer in will AI replace designers?

Conclusion

AI now sits in the graphic designer's toolbox next to Photoshop and Illustrator, not above them. Use it where it is strong: options, moodboards, clean-up, scenes, motion, voice and small tools. Keep your judgement where it is weak: the idea, Arabic type, the brand system, the client's context and the ethics of what you publish. Direct it, check it, and be open about it.

If you need a brand, packaging or a film made with AI as a tool and a designer's judgement in charge, tell me about your project.