AI for Freelance Designers: Faster Concepts Without Rights Issues

Use AI for exploration, such as moodboards, thumbnails, colour and layout variations, and draw or rebuild the final artwork yourself. Pick generators whose training and terms you can defend, never prompt for a living artist's style or a brand's look, keep a record of your process, and tell the client in writing how AI was used.

The two risks pull in opposite directions. One is that AI output resembles someone else's work closely enough to infringe it. The other is that pure AI output may have little or no copyright protection, so a client who pays for a logo could find they can't stop a competitor copying it. Human-made finals deal with both.

Follow me on Instagram@sagnikteaches

Where AI helps a design job, and where it adds risk

Split a typical identity or packaging job into stages and the picture gets clearer. Most of the time AI saves is early, where the rights risk is lowest.

Connect on LinkedInSagnik Bhattacharya

If you only change one habit, make it the last two rows of the "what stays human" column: finals are drawn, and production edits never change what the product looks like.

Subscribe on YouTube@codingliquids

Choosing a generator by its terms, not just its output

The best-looking output isn't the deciding factor for client work. Before a tool goes into your process, check four things in its terms and help pages: what it was trained on, who owns the output, whether your images are public, and what protection (if any) comes with a paid plan.

• Adobe Firefly. Adobe says its own Firefly models are trained on licensed content such as Adobe Stock, openly licensed material and public-domain content, and markets them as commercially safe. Two cautions. The Firefly app now also offers partner models from other companies, and Adobe's training claims cover its own models, not those; the app asks you to confirm when you pick a non-Adobe model, and Content Credentials record which model made an image. And the contractual IP indemnity Adobe offers is an entitlement for enterprise customers, not something an individual Creative Cloud subscription includes.
• Midjourney. Its terms say you own what you create, with a condition: a company, or an employee of a company, with more than $1 million a year in revenue must be on the Pro or Mega plan to own its assets. Images are visible in its public gallery unless you use Stealth mode, which comes with Pro and Mega. For unreleased client work, that visibility matters. Whether the subscription is worth it is covered in is Midjourney worth it for marketing images.
• Any other tool. Look for the same answers. If the terms are vague about ownership, or say the vendor can use your outputs publicly, keep it to private exploration.

A concept workflow that keeps you the author

The aim is a process where AI widens your options and every decision that shapes the final design is yours, and where you could show that if asked. Seven steps:

1. Write the design brief summary yourself (20 minutes). Three to five sentences on audience, feeling and constraints. This becomes the yardstick for everything the AI produces.
2. Generate wide (45-60 minutes). 40 to 80 exploratory images across moods, palettes and compositions. Quantity is the point; don't polish anything.
3. Select and annotate (30 minutes). Pick eight to twelve, and write one line on each saying what you're taking from it: "warm timber tone", "asymmetric crop", not "this one".
4. Sketch the routes by hand or in vector (two to four hours). Build two or three concept routes yourself from your annotations. Don't trace AI images.
5. Present honestly. Show AI moodboards labelled as such, next to your routes, so the client sees the thinking and knows what's generated.
6. Build the finals from scratch. Final lettering, marks and illustrations are drawn in your usual tools. Mock-ups can use AI scenes.
7. Keep a process log as you go. It takes five minutes a stage and is your evidence of authorship.

Step 3 is the one people rush, and the annotations are what turn a pile of generations into your own design thinking. Three lines from the furniture workshop job further down, illustrative:

MB-07  keep: raking side light across end grain; ignore the chair
MB-19  keep: warm oiled-oak tone against cold grey concrete; not the layout
MB-33  keep: the gap between two boards as a negative-space idea

Each line says what you're taking and, just as usefully, what you're leaving. MB-33 later fed the dovetail monogram route, and that chain from annotation to route is exactly what the log needs to show if anyone asks who designed the mark.

A filled-in log for one job looks like this:

PROCESS LOG - [client], identity refresh             Designer: [initials]
Stage          Date     Tool / model                   What I did / kept
Moodboard      03 Mar   Firefly Image (Adobe model)    62 images; kept 10;
                                                        notes in moodboard.pdf
Moodboard      03 Mar   Midjourney (Stealth on)        24 images; kept 2
                                                        (lighting refs only)
Routes         05 Mar   Pencil, then Illustrator       3 routes drawn from notes;
                                                        no AI images traced
Wordmark       09 Mar   Illustrator                    Drawn by hand from route B;
                                                        letterforms built, not type
Mock-ups       10 Mar   Firefly Image + Photoshop      Workshop scene backgrounds
                                                        generated; logo placed by me
Delivered      14 Mar   -                               Finals: no AI-generated elements
                                                        Mock-ups: AI backgrounds, noted

Prompting for ideas without borrowing someone's style

Resemblance problems usually start in the prompt. "In the style of [living illustrator]" or "like [famous brand]'s packaging" asks the model to reproduce what makes someone else's work distinctive. Describe the qualities instead: line weight, era, materials, colour temperature, texture, mood.

A before and after for one moodboard prompt, illustrative:

Before: botanical label illustration in the style of [named living
        illustrator], for a gin bottle

After:  botanical label illustration, fine single-weight ink line, loose
        cross-hatching, off-white uncoated paper texture, two-colour print
        look in deep green and copper, plants drawn from life rather than
        stylised, generous margins

The second prompt takes a minute longer to write and gives you more control, because each quality is one you chose and can change. It also gives your process log an entry you'd be comfortable for a client, or that illustrator, to read.

A text chatbot is useful before you open an image tool, to widen the concept directions. For an illustrative food truck selling slow-cooked street food, a prompt might be:

I'm designing a visual identity for a food truck selling slow-cooked street
food at markets and events. Audience: office workers at lunch and weekend
market crowds. It must read from 10 metres away on a truck side and work on
a small paper tray. Suggest 5 distinct concept directions. For each: a name,
the visual idea in two sentences, a colour approach, a type approach, and one
risk. Don't reference existing brands or named artists.

The reply might look like this (illustrative):

1. "Low and Slow" - Heavy, rounded letterforms that look like they've melted
   slightly, with a single flame-orange on charcoal. Type: custom bold rounded
   sans. Risk: melting type is common in street food and can look generic.
2. "Market Ticket" - The identity as a printed raffle ticket: perforated edges,
   serial numbers used as daily specials. Colours: newsprint cream, red, black.
   Risk: detail lost at small sizes.
3. "Smokehouse Sign" - Hand-painted sign-writer lettering with drop shadows,
   warm golds on deep green. Risk: close to a common barbecue look.
4. "Clock Face" - The hours of cooking as the core motif: a stylised clock
   with hands as a fork and knife. Risk: fork-and-knife clocks are widely used.
5. "Steam Lines" - Minimal: three hand-drawn steam curves as the mark, type in
   a narrow condensed face. Risk: may feel too quiet on a busy market row.

How to use it: the AI flagged its own clichés fairly on 1, 3 and 4, which is useful; drop 4 entirely, because that motif is widespread enough to create resemblance trouble and weak distinctiveness. Route 2 is the most ownable idea but its risk note is too mild: test it at tray size before you invest in it. Then take the two or three strongest into your image tool with prompts describing qualities, not references.

Rebranding a two-person furniture workshop, stage by stage

The client in this illustration is a two-person furniture workshop that makes solid-wood tables and chairs to order and wants a new identity for its website, workshop sign and burnt-in maker's mark on each piece. The freelance designer quotes a fixed fee for three routes, one developed identity and a small asset pack.

• Before AI, the designer's exploration stage (research, moodboards, rough routes) took around ten hours on jobs like this.
• With AI, the moodboards come from about 60 Firefly generations of workshop scenes, wood grain, joinery details and lighting, plus a chatbot pass on concept directions. Exploration drops to about four hours. The time saved goes into drawing the routes more carefully.
• The routes are drawn by hand: a monogram based on a dovetail joint, a wordmark with letterforms cut like chisel strokes, and a stamp-style mark that works as a branding iron.
• The final is the stamp mark, drawn in vector from the designer's sketch and tested at 25 mm for the branding iron. No AI-generated shapes are in it.
• The handover includes the log, the finals, and mock-ups of the website and sign with AI-generated workshop backgrounds, marked as such in the delivery note.

The client ends up with a mark whose authorship is clear, delivered faster, with the exploration visible and honestly labelled. The designer kept the fee and gave back a shorter timeline rather than a discount, which is a pricing choice worth making deliberately.

The sum behind that choice, with illustrative figures: a fixed fee of $2,400 for a job that used to take 30 hours works out at $80 an hour. Take six hours out of exploration and the same fee earns $100 an hour for 24 hours' work. Cutting the fee by the time saved would hand the whole gain to the client; keeping the fee and delivering a week sooner shares it, because the client gets their new identity earlier.

One production edit on this job nearly went wrong. For a wide website banner, the designer extended a photo of a finished dining table with generative fill. At a glance the extension looked right, but the fill had continued the tabletop with a rounded edge where the real table has a square one, and added a stretcher bar between the legs that the design doesn't have. A customer ordering from that photo would have expected a different table. The designer rebuilt the banner from the original photo and kept generative fill for the wall and floor behind the table, where nothing is for sale.

When the client arrives with AI images of their own

More briefs now come with a folder of images the client generated themselves and the words "something like this". Treat them exactly as you'd treat a moodboard from a competitor's website: a statement of taste, not a starting file. Ask three questions before you open them in your design software. Which tool made them, and on what plan? (If it was a public gallery tool without a private mode, those images are already visible to anyone.) Did the prompt name an artist, a brand or a film? And which parts does the client actually like: the colours, the layout, the mood, or the specific object in the middle?

A small bakery, as an illustration, sends nine images made on a basic plan of a generator with a public gallery, and one of the prompts names a famous animation studio. Asked what they like, the owner points at the flour-dusted morning light and the hand-lettered chalkboard that appear in three of them, and shrugs at the cartoon characters. So the images are already public (don't file them as confidential), the studio's name goes nowhere near your prompts or your log, and the useful content of the whole folder fits in one line: "soft morning light, flour texture, chalk lettering warmth, no characters".

Then write the answer into your brief summary in your own words ("deep greens and brass, a lot of empty space, feels like a gentlemen's outfitter from the 1920s") and work from that. Never trace, recolour or vectorise the client's AI image as the final, even when they ask you to: they'd be paying you for a design with the weakest possible ownership position and none of your authorship. If they insist, put it in writing that the final is derived from their AI-generated reference and that your usual assurance about hand-drawn finals doesn't apply.

What to tell the client, and how to word it

Put a line in your terms, then a specific note in each delivery. Contract wording that suits most freelance design work:

I may use AI image and text tools for research, moodboards and presentation
mock-ups. Final logos, lettering and illustrations are drawn by me and contain
no AI-generated elements unless we agree otherwise in writing. I'll tell you in
the delivery note where AI-generated material appears in anything I supply.

And a filled-in delivery note line: "Mock-ups 1-4 use AI-generated workshop backgrounds (Adobe Firefly); the logo, wordmark and maker's mark are drawn by hand and contain no AI-generated elements." For the wider question of whether and how to tell clients, see should freelancers tell clients they use AI. If you work through an agency, its contract may set the rules for you; AI clauses for agency contracts shows the kind of terms to expect.

Checks before you hand over the files

• Reverse image search any AI-influenced concept that made it into a route, to see whether it lands close to existing marks or illustrations.
• Trademark search. Before the client commits, search the trademark registers where they trade, or ask their lawyer to. Looking unlike other logos to you isn't the test.
• Artefacts. Check mock-ups for AI tells: garbled text on background signs, extra fingers, impossible joinery, reflections that don't match. The full routine is in how to check AI-generated images for errors.
• Fonts and stock. Confirm licences for any type or stock used alongside AI material; these trip up more handovers than AI does.
• Product accuracy. If a mock-up shows the client's products, make sure they're shown as they are.

A check that finds something is the check working. On the workshop job, say the reverse image search on the dovetail monogram route turns up a woodworking school's logo built from the same interlocking tails at a similar angle. It isn't a copy, since the route was drawn from the designer's own notes, but it's close enough that the client would struggle to call the mark distinctive. Record the search and the result in the log and raise it when you present the routes; in this job it was one reason the stamp mark won. Finding that logo after the branding iron has been made would cost far more than an hour of redrawing.

Prompts are where it usually goes wrong. In an illustrative but plausible case, a designer produced a character illustration for a subscription box company's packaging by lightly refining an AI image generated with an illustrator's name in the prompt. It passed the client's review. After launch, the illustrator's followers spotted the likeness on social media and the client had to withdraw the design and pay for a redraw. The prompt was the mistake, and the process log would have shown it. For logo-specific risks from the client's side, should you design your logo with AI is worth sending to clients who ask for "just an AI logo".

Designers' follow-up questions

Can I use AI to clean up or vectorise my own sketches?

Using AI to trace, tidy or recolour work you drew yourself is a different case from generating a design from a prompt, because the creative decisions are already yours. Keep the original sketch in your process log, check the result hasn't added shapes you didn't draw, and review every curve before delivery. Auto-traced vectors are often messy enough to need rebuilding anyway.

Is generative fill on a client's photo a problem?

Small edits such as removing a cable or extending a background are routine, but know which model did the work, since Adobe now offers partner models alongside its own Firefly models in some apps. Tell the client if a substantial part of an image is generated, and never generate or alter the client's actual product, because the image must still show what customers will receive.

Should I keep Content Credentials attached to files I deliver?

Content Credentials record whether generative AI was used and, in Adobe's apps, which model. Keeping them on concept images is honest and helps the client if anyone asks later. For final artwork you drew yourself there may be nothing to record. Don't strip credentials to hide AI use, because that undoes the transparency your contract promises.

Further reads

• Can You Legally Use AI-Generated Images in Your Marketing? — The legal questions clients ask about AI images in their marketing.
• Brand Guidelines for AI Images: Colours, Style and Limits — Writing brand rules so AI imagery stays on-brand later.
• How Freelancers Use AI for Invoices, Proposals and Chasers — Saving time on the admin around each design job.
• Should You Charge Clients Less When AI Speeds Up Your Work? — Pricing when AI makes the exploration stage faster.
• How to Label AI-Generated Images and Video on Social Media — When clients need to label AI imagery they publish.
• Copyright and Client Data Risks of AI in Architecture Practices — How another design profession handles the same rights risks.
• How Interior Designers Use AI for Mood Boards and Presentations — How interior designers use AI for concept boards, real-product selection boards and client decks, with prompts, sample outputs and the traps to avoid.
• AI Room Visualisation From Client Photos: A Designer's Workflow — Turn a client's room photo into an honest concept visual: shooting rules, a lock-list prompt, drift checks and the caption every image needs.
• Can You Sell an Online Course Made With AI? — Udemy's AI rules, who owns AI-made lessons, three example courses with different verdicts, and disclosure wording you can copy.
• Do You Own the AI-Generated Content Your Business Publishes? — What AI vendors' terms really hand you, why copyright can still be thin, and the edits, checks and contract wording that make published AI content safer.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Adobe Firefly business pages and help on partner models and Content Credentials; Firefly legal FAQs for enterprise customers; Midjourney Terms of Service; Copyright and Artificial Intelligence Part 2 report on copyrightability.