Answer in brief
An honest AI portfolio entry separates the human brief and decisions from the generation and the manual refinement. Midjourney's and OpenAI's terms differ, Dribbble sets its own rule, and none of it guarantees legal cleanliness automatically.
Four categories of contribution, checked against the record
This piece works through one clearly labeled teaching example that splits a project's contribution into four honest categories — the brief, the author's decisions, the tool's output and the manual selection and refinement — and checks each category not against general claims about AI, but against the exact wording of two different image generators' terms of use, a standing rule at a major portfolio platform, and the official position of the U.S. copyright regulator.
A worked example: rebranding the fictional coffee shop “Noon”
An honest description of an AI-assisted portfolio project rests on four separate points rather than one line: what brief and completion criteria the author set before opening any tool; what style, selection and composition decisions the author made; what the image generator actually did and under which service's terms; and what was selected from the output and refined by hand. What makes this work is specificity at each step — the tool's name and version, the share of the work that was generated, the amount of manual refinement — not the blanket phrase “made with AI” and not silence about the tool at all.
The rest of this piece shows that method on a teaching scaffold, not on an account of a real commission: VJOURNAL's editorial team did not run or commission this project, it has no fictitious author, client or measurable outcome — it is a hypothetical frame meant to demonstrate a method that applies to any real project a reader might have. Picture a beginning designer preparing a rebrand of a fictional coffee shop called “Noon” for their portfolio: a logo, a palette, a set of menu cards and a social-media cover — and at one stage using an image generator to quickly produce visual options for a packaging texture. Below, each stage of this hypothetical project is broken down into what the person did and what the tool did.
The brief and constraints an author sets before opening a generator
The first category of an honest description is what, in the “Noon” example, the author defines before opening any tool: the brief. The designer sets the audience (neighbors in the area, not tourist foot traffic), three to five words describing the brand's character (“warm,” “unhurried,” “not overly sweet”), constraints (the logo must read equally well on a cup, a sign and a social avatar) and a completion criterion (recognizable without a caption across three different surfaces). In this example a person writes the brief — and that is exactly what should be stated first in a project description, rather than opening the account with the generator. If a language model, rather than only an image generator, helped write the brief, the audience description or the portfolio copy itself, that contribution belongs in the disclosure too: it covers any material AI assistance, not only a generated image.
Decisions a person made in the project, not the generator
The second category is the decisions an author makes before, during and after working with a tool: which references to gather and why those specifically, which palette to narrow down from a broader set, which composition to pick from several options, what generated material to use as a base and what to discard entirely. In the “Noon” example, the designer writes the prompt to the generator personally (describing exactly what the packaging texture should look like — “warm” paper, a specific tone, a reference to morning light) rather than simply typing the brand name and waiting for a finished result: choosing that wording is itself an authorial decision, but it does not substitute for or guarantee the artistic value of the result — that still depends on what the particular generator produces. This decision should not be confused with a basis for copyright in the generated result itself: whether prompt wording alone is enough to make the user the author of the image is a question the U.S. Copyright Office's newer report on the copyrightability of generative-AI output, published January 29, 2025, addresses separately. Per the Office's press release about that part of the report, protection can attach to a human's creative arrangement or modification of a generator's output, “but not the mere provision of prompts”; prompt wording alone is not put forward here as a basis for authorship.
Generator terms: what an author gets along with the image
The third category is the work of the tool itself, and an honest description here calls not for the blanket phrase “with the help of AI” but for understanding what a particular generator actually offers and under what terms, according to the maker's own documents — because the terms differ noticeably between tools.
Midjourney's terms: rights to the output depend on the plan, and so does privacy
Under Midjourney's terms of service (the version in force since May 27, 2026), rights to the Assets a user creates are assigned to that user to the broadest extent permitted by applicable law. But the rule has a carve-out: a company earning more than $1 million a year in revenue — or its employee — must hold a Pro or Mega subscription to own the results of its own generation. By default, created images are public and available to other users for remixing; a Stealth mode can make them private, but per the service's own documentation this is not a separately purchasable add-on but a feature available only on the Pro and Mega tiers, and even there Midjourney commits only to “best efforts” not to publish such images — that is, it does not guarantee the result. Under the same terms, anything created in the service's shared spaces — the Discord channel, for instance — is visible to other participants regardless of whether Stealth mode is on. Per the service's documentation, Midjourney has no free tier: it runs on four paid subscriptions — Basic, Standard, Pro and Mega. For the teaching example, this means: had the “Noon” designer used Midjourney on a Basic or Standard plan, the generated packaging texture would by default have been visible to other users of the service, with Stealth mode unavailable — it would not have stayed a private draft.
OpenAI's terms: output is assigned to the user, but uniqueness is not guaranteed
OpenAI's Terms of Use (published by OpenAI OpCo, LLC; in force since January 1, 2026) cover use of ChatGPT, DALL·E and other OpenAI services in the version for individuals: users in the European Economic Area, Switzerland and the United Kingdom are covered by a separate version of the terms, and use of ChatGPT Enterprise and the API is governed by separate Business Terms. Under these terms, as between the user and OpenAI and to the extent permitted by applicable law, Output is assigned to the user — the company transfers whatever rights it may hold in the Output. But the same terms separately note that Output may not be unique and that another user could receive similar output from the service — formal ownership does not mean a similar image will not turn up somewhere else. Elsewhere, the terms explicitly bar a user from passing off a generator's output as human-made when it is not — that prohibition limits a false claim about the output's origin, but on its own it does not require disclosing that AI was used; that recommendation, for instance, is made separately by the portfolio platform Dribbble.
Selection and manual refinement: what generated material reached the final submission
The fourth category is what happens after generation: from several texture options, the author in the teaching example picks one, crops and recomposes it for the required menu-card format, manually matches its color to the rest of the brand palette, and adds typography and layout that the image generator never produced at all. This is the category most often lost in the phrase “AI designed it”: in this teaching example, a person — not the generator — handles the typography, the final layout composition and the color matching. The portfolio platform Dribbble states its rule without qualification: “Do not share content that is wholly generated by AI.” The ban applies whether or not the tool's involvement is disclosed in the description; separately, for work where AI participated alongside a person, the platform recommends noting the tool's contribution in the work's description. In other words, Dribbble's rule is not about whether a disclosure label makes publishing a wholly generated work acceptable — it cannot be published either way; the point is that a tool's involvement in a partly authored work is worth mentioning.
How to state a tool's contribution in a portfolio description
The U.S. Copyright Office requires applicants to disclose the presence of AI-generated content in a work submitted for registration and to briefly describe specifically the human author's contribution — that is a requirement for legal registration in the United States, not for portfolio copy, but the principle carries over: state not the bare fact that AI was used, but exactly what the person did. A working formulation for a project description is neither “designed with the help of AI” nor silence about the tool, but detail on each of the four steps: what brief the author set, what decisions the author made personally, which generator and version were used and for which element, and what was selected from the output and refined by hand. That formulation satisfies Dribbble's recommendation to disclose AI's contribution while also not passing the generator's output off as entirely manual work, which OpenAI's terms of use separately prohibit.
For “Noon,” such a description might read (the text is hypothetical, tied to the fictional example rather than a real project — a sample formulation, not an account of a commission that took place): the “Noon” coffee-shop rebrand is a teaching project. The author set the brief and completion criteria independently: audience — neighbors in the area; brand character — warm and unhurried; requirement — the logo must read equally well on a cup, a sign and a social avatar. References, palette and menu-card composition were chosen and arranged by the author. The packaging texture for one menu card was generated in Midjourney (Terms of Service version of May 27, 2026; plan assumed to be Standard) from the author's own text prompt describing the paper, tone and light; one variant was selected from several. Further work — cropping to the card format, matching color to the brand palette, typography and final layout — was done by hand in vector and raster editors; the generator produced none of it. A weaker version of the same description — “designed with the help of AI,” or no mention of the tool at all — undervalues the part a person did entirely and leaves the viewer unable to tell what was generated from what was not; a stronger version names the tool, the plan, the specific element and the amount of manual work, without apologizing for using a generator or crediting it with what it did not do.
What a generator does not guarantee: copyright and third-party claims
None of the sources reviewed supports a promise that any generated image is legally clean for any use. Under the official position of the U.S. Copyright Office — a single country's regulator, not a rule shared across every jurisdiction — only material created by human authorship can be protected by U.S. copyright, and an applicant should explicitly disclaim AI-generated content above a de minimis level in a registration application: the part of a project generated by a tool does not itself receive copyright protection in the United States, unlike the human contribution around it. In other jurisdictions, including Russia, the answer may differ — this article makes no such comparison. Separately, the tools themselves do not guarantee freedom from third-party claims: Midjourney provides its service and outputs “as is,” and OpenAI provides its services “as is”; both explicitly disclaim any warranty of non-infringement. The practical conclusion for a portfolio follows from this: naming the exact tool and version used for a given element matters not just as a courtesy to a platform, but because the real rights to the result depend on the tool and its current terms — and, as the difference between Midjourney and OpenAI above shows, those terms differ between generators and can change when the terms are updated.
Content Credentials: a technical record of a file's origin, where a tool supports it
Beyond a written description in a portfolio, there is also a technical way to record a file's origin: the Content Credentials standard maintained by the Coalition for Content Provenance and Authenticity (C2PA); per the project's official site, more than 500 companies participate in the work, among them Microsoft, Adobe, Intel, BBC, Sony, OpenAI, Google, Meta and Amazon. The standard attaches a verifiable record to a file of how and with what it was created and edited. It is neither universal nor mandatory — not every generator and not every portfolio platform supports it today, and it cannot yet be relied on as the sole proof of a work's origin — but where a tool does export such a mark, it can supplement a written description of contribution rather than replace it.
Practical checklist
- State the brief and the completion criteria the author set personally in the project description — before mentioning the tool.
- Name the specific image generator and, if known, the version or plan — not the blanket phrase “with the help of AI.”
- List separately what a person decided and made: references, composition, typography, final layout, color matching.
- Check the terms of use in force for that specific tool at the time of publication — wording on rights to output changes periodically.
- Do not publish content wholly created by AI on Dribbble, labeled or not; for work with AI involvement, the platform recommends stating the tool's contribution in the description. Check other platforms' rules separately — they differ.
- If a platform or file exporter supports Content Credentials, keep that record with the file rather than relying on the written description alone.
Questions and answers
Is it necessary to name the specific generator in a portfolio, or is a general phrase like “made using AI” enough?
Comparing the terms of use shows specificity is more useful than a general phrase: different generators carry different terms on rights to output, and without the tool's name a reader of the description cannot judge exactly what rights the author holds over the work shown.
Can a project be registered for U.S. copyright if part of the image was generated by AI?
Under the U.S. Copyright Office's position, protection covers only the human contribution, and any AI-generated part above a de minimis level must be explicitly disclaimed in a registration application: the whole project cannot be submitted “as is,” and the Office reviews the human part (for instance, refinement and layout of the image) separately, not always finding it eligible for protection. For other countries, including Russia, this article gives no answer: their registration rules are their own.
Should a language model's help with the brief or the project description text be disclosed too, not only an image generator's contribution?
Yes: disclosure, as discussed here, covers any material AI assistance that affected the finished work or its description, not only a generated image — otherwise the description of contribution ends up selectively honest.
What if a specific portfolio platform does not require a label for AI use?
A platform's lack of such a requirement does not cancel the tool's own terms of use: OpenAI, for instance, prohibits a user outright from passing off output as human-made when it is not, regardless of platform rules. Beyond that, an honest description of contribution protects the author: if someone later notices a mismatch between the claimed and actual amount of manual work, trust in the rest of the portfolio suffers more than it would have from disclosing the tool's role from the start.