Adobe Firefly vs Midjourney: Which Is Safer for Client Images?

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Adobe Firefly vs Midjourney: Which Is Safer for Client Images?
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Adobe Firefly vs Midjourney: Which Is Safer for Client Images?

Adobe Firefly is the safer default for client marketing images: Adobe says outputs from its own Firefly models are safe for commercial use and attaches Content Credentials. Midjourney lets you own what you make but supplies images "as is", with no non-infringement warranty, and firms earning over $1 million a year need its Pro or Mega plan.

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

Neither tool makes an image safe by itself, though. Adobe's statement covers its Firefly models, not the partner models from Google, OpenAI and others that sit in the same app. Midjourney has the deeper set of style tools and, at volume, a low cost per image, but its images are public by default unless you pay for Stealth Mode. So the useful answer is a split: Firefly for deliverables that carry the client's name, Midjourney for concept work under rules you agree with the client, and a written record for both.

What does "safe for clients" have to cover?

When a client publishes an image you made, "safe" means five separate questions, and Firefly and Midjourney score differently on each.

  1. Permission to use. Do the vendor's terms let you and your client use the output commercially, and does any plan or revenue condition apply?
  2. Provenance. What was the model trained on, and does the vendor stand behind the output if someone claims it infringes their work?
  3. Confidentiality. Can strangers see the images, or the client's unreleased campaign, while you're making them?
  4. Your promise. What does your contract tell the client, and does any of the risk sit with the vendor rather than you?
  5. Disclosure. Can the client show, if asked, that an image is AI-generated and how it was made?

A tool can pass one question and fail another. Midjourney gives you ownership (question 1) but makes no promise about infringement (question 2). Firefly answers question 2 better for its own models but offers indemnity only to eligible enterprise customers (question 4). The sections below go through each with the vendors' own wording.

How do Firefly and Midjourney terms differ for client work?

Adobe's terms cover Firefly's own models for commercial use, while Midjourney's give you ownership but no warranty and make images public unless you pay for Stealth Mode.

Here are the points that matter for client work, taken from Adobe's Firefly plans page and credits FAQ and from Midjourney's Terms of Service (version effective 27 May 2026).

QuestionAdobe FireflyMidjourney
Commercial useOutputs from Firefly models can be used commercially; beta features can be too unless the product says otherwiseYou own what you create "to the fullest extent possible under applicable law"
ConditionsPartner models: "you're responsible" for deciding if the output suits your projectCompanies with over $1,000,000 a year in revenue must be on Pro or Mega to own their assets; upscales of other people's images stay theirs
Training dataFirefly models trained on licensed content and public-domain content where copyright has expiredNot stated in the terms
Vendor assuranceA statement that Firefly-model outputs are safe for commercial use (not an indemnity)Services and assets provided "as is", with no warranty of non-infringement
IndemnityEligible enterprise customers only, for select outputs; beta outputs excludedThe other way round: you indemnify Midjourney against third-party claims
Who can see your workPublic only if you submit it to the Firefly community gallery; Adobe says your content isn't used to train its generative modelsPublic and remixable by default; Stealth on Pro and Mega uses "best efforts" not to publish
Licence back to vendorNot covered in the pages used hereYou grant Midjourney a perpetual, worldwide licence to your inputs and outputs
LabellingContent Credentials attached automaticallyNot stated in the terms

Two rows deserve a second look. The licence-back row means anything you feed Midjourney, including a client's photo used as a reference, is covered by a broad licence to Midjourney, which some clients won't accept for unreleased material. And the confidentiality row has a trap even on Stealth: Midjourney's terms say images made in a shared or open space, such as a Discord channel, are visible to anyone in that channel regardless of Stealth. Generate client work in your own web account or a private message, not a community channel. More on how designers handle these terms is in Midjourney for graphic designers: client work and licences, and the Adobe side is in what designers can promise clients about Firefly.

Which tool matches a client's look better?

Midjourney has the deeper style toolkit, while Firefly is the more controllable tool for matching a client's look.

In Midjourney, a Style Reference captures the colours, medium, texture and lighting of an image you supply, without copying its objects or people, and works on version 6 and later. The --sw parameter sets how strongly it applies, from 0 to 1,000 with 100 as the default. Instead of an image you can use a numeric style code, browse codes in the Style Explorer, or build your own with the Style Creator, and moodboards and personalisation profiles tune the look further. The current default model, V8.2, has been the default since 24 July 2026, and its Edit Model replaced the older Omni Reference and Character Reference tools, so older tutorials that rely on those are out of date.

In Firefly, the equivalent is the style reference on Generate image, with a Strength slider that Adobe documents for its Firefly Image models, plus effects for colour tone, lighting and camera angle. Firefly's custom models, still in beta, can be trained on a client's own images to reproduce a style or subject, at 500 credits per training and 20 per generation on Adobe's rate card. Adobe lists them for Creative Cloud individual plans, Creative Cloud Pro and Pro Plus for teams, and some enterprise plans, so check eligibility before you quote a client for it.

One rule applies to both: you need the right to use the reference image. Midjourney's terms require you to have the permissions for content you upload to its Editor, Adobe's community gallery guidance says you must own a reference file or have the rightsholder's permission, and a style reference lifted from another brand's campaign is exactly the kind of input that makes an output unsafe however clean it looks. Use the client's own photography, or images you've licensed, as references. There's a full method for one campaign look in Midjourney Style Reference across a campaign.

How do you work out the cost per usable image?

The cost per usable image is the plan price divided by the images you actually deliver, which depends on how many attempts each keeper takes. The inputs, from each vendor's pages:

  • Firefly: Standard is $9.99 a month with 2,000 credits; Pro $19.99 with 4,000. Standard image generations with Firefly models use no credits on a paid plan. Firefly Image 5 costs 10 credits a generation; partner models cost more, such as 20 credits for GPT Image at medium quality on Adobe's current limited-time rate.
  • Midjourney: Basic $10 a month (200 Fast minutes, no Relax, no Stealth), Standard $30 (15 Fast hours, unlimited Relax images), Pro $60 (30 Fast hours, Stealth), Mega $120 (60 Fast hours, Stealth). Annual billing takes 20% off. Midjourney says an image prompt usually takes about a minute of GPU time: 0.8 minutes for a standard-definition prompt and 1.3 for HD. Extra Fast time is $4 an hour. Each prompt returns a grid of four options.

So the formula is the same for both: monthly plan cost ÷ keepers delivered, with a check that the attempts fit the allowance. For a deeper look at Midjourney's side of this, see what a usable Midjourney image really costs.

If you run the cost-per-image sums in a spreadsheet, The Complete Excel Course with AI Integration on Coding Liquids covers that side.

A landscaping client's month, priced both ways

Here's the main worked example. An illustrative two-person design studio makes marketing images for a landscaping company: 24 finished images a month for its website and social channels, such as seasonal planting scenes, a patio at dusk for a lighting service, a tidy lawn edge for a maintenance plan. The studio's revenue is well under $1 million, but the client wants its spring campaign kept private until launch.

From a trial week, the studio estimated four Firefly generations per keeper and two and a half Midjourney prompts per keeper, each prompt giving four options. Those ratios are illustrative; measure your own over a week before relying on them.

Firefly StandardMidjourney BasicMidjourney Pro
Monthly price$9.99$10$60
Attempts for 24 keepers96 generations60 prompts (240 options)60 prompts (240 options)
Allowance used0 credits with a standard model; 960 of 2,000 with Firefly Image 5About 48 to 78 of 200 Fast minutes (SD to HD)About 48 to 78 of 1,800 Fast minutes (SD to HD)
Cost per keeperAbout $0.42About $0.42$2.50 ($2.00 on annual billing)
Private while in progress?Yes: public only if you choose Submit to galleryNo: public by defaultYes, with Stealth
Vendor stands behind the output?Commercial-use statement for Firefly modelsNo warrantyNo warranty

The sum makes the trade-off plain. On price alone, Firefly Standard and Midjourney Basic come out the same at about $0.42 a keeper. But Basic fails the client's confidentiality requirement, so the real Midjourney comparison is Pro at $2.50 a keeper, six times Firefly's figure at this volume. Pro only gets cheap per image when you use its allowance: at 240 keepers a month the same ratios would use up to about 780 Fast minutes at HD, still inside Pro's 1,800, and the cost per keeper would fall to $0.25.

Review time matters too. Four Midjourney options per prompt means the studio looked at 240 images to find 24, against 96 in Firefly. At an illustrative 20 seconds a look, that's 80 minutes against 32 a month. Not a deal-breaker, but it belongs in the price you quote.

The studio's decision: Firefly for every image the landscaping company publishes, using a standard Firefly model and the client's own garden photos as style references. Midjourney Pro stays on for two other clients who want illustrated campaign concepts, where its style codes earn their keep and the client has agreed in writing to the terms above.

The patio image, prompted in each tool

To show how the working styles differ, here is one of the landscaping images made both ways. In Firefly, with a standard Firefly Image model selected, the client's own photo of a finished patio attached as the style reference and Strength at about two-thirds, the prompt describes the scene and leaves the look to the reference:

Sandstone patio at dusk with warm low-level path lights, a small olive tree
in a large grey planter, two simple outdoor chairs, lawn edge in the
foreground, calm and natural, no people, no text, no logos

A sample result, described for illustration: the tones matched the client's photo well, but the first batch put a garden lantern with a maker's mark on its base, which went through Generative fill. One image had path lights floating above the paving and was rejected. The keeper went into the delivery log with its Content Credentials intact.

In Midjourney, working in the studio's own web account with Stealth on, and with a licensed stock photo as the reference rather than the client's own picture (because of the licence Midjourney's terms take over inputs), the same scene uses parameters to steer the look and strip out marks:

sandstone patio at dusk, warm low-level path lights, small olive tree in a
large grey planter, two simple outdoor chairs, lawn edge in the foreground,
calm natural garden photography --ar 3:2 --sref [licensed reference URL]
--sw 150 --no text, logo, lettering, people

A sample result, again illustrative: the grid of four had more atmosphere than Firefly's, and one option was striking enough to make the client's designer ask for it. But two options added a brick wall the client doesn't build, and the striking one showed a planting style the firm doesn't offer. That's the real risk with a more creative model on client work: an image can be beautiful and still promise a service the business can't deliver. The studio kept Firefly's version for the website and saved the Midjourney one for an internal ideas board.

Two habits came out of this test. First, every prompt for a client image ends with an exclusion list, because invented marks were the most frequent reason a keeper failed the studio's final check. Second, whoever approves the image checks it against the client's real service list, not just against the brand look. A sample note the studio now sends with each batch reads: "Hi [first name], here are this month's 24 images. All were made with Adobe Firefly's own models and checked for logos, lettering and services you don't offer. The delivery log is attached. Reply with any you'd like changed before Friday."

When should you use Midjourney and when Firefly?

Use Midjourney for concept work and Firefly's own models, with a full record behind them, for finished deliverables, because the closer an image gets to the public the safer it needs to be. Here's how that played out across the studio's work, as four short examples.

  • Pitch mood board for a new client. Midjourney on Pro with Stealth, because the style range helps sell a direction and the board never leaves the meeting. The board is labelled "AI concept images, not for publication".
  • Website hero image for the landscaping company. Firefly, standard model, client photo as the style reference, Content Credentials left attached.
  • "Before and after" garden post. Neither tool. An AI image shown as a finished project misleads customers about work the firm has actually done, so before-and-after posts use real photos only. AI images can illustrate an idea ("a low-maintenance front garden could look like this") if the post says so.
  • Seasonal tips carousel. Firefly for the backgrounds, with the text and logo added in a layout tool, so nothing in the image itself carries words or marks.

The before-and-after example can do the most damage of the four, and it has nothing to do with copyright. A client-safe image is one that doesn't mislead the client's customers, whatever tool made it.

A contract clause and a delivery log you can adapt

Writing your process down protects you more than any tool choice. Here's a clause a studio might put in its terms, to adapt and have a qualified adviser review for your situation:

AI-generated images. Some images we supply are created or edited with
generative AI tools. We will tell you which ones on each delivery note.
Unless agreed otherwise, published images are made with Adobe Firefly's own
models, which Adobe states are suitable for commercial use. Concept and
mood-board images may use other tools and are not licensed for publication.
We do not use your unpublished material as a reference in any tool whose
terms grant the tool provider a licence to it, without your written consent.
AI-generated images are not presented as photographs of real work.

And here's the studio's delivery log for one month, filled in for illustration. One row per delivered image is enough:

ImageTool and modelPlan, privacyReference usedChecks passedUse
LS-0412 patio at duskFirefly, standard Firefly Image modelStandardClient's own patio photoZoom check, no marks, Content Credentials presentWebsite
LS-0413 lawn edgeFirefly, Firefly Image 5Standard, 10 creditsNoneZoom check, no marksSocial
CB-0108 board tile 3Midjourney V8.2, style codePro, Stealth onStudio's licensed photoConcept onlyPitch board

When a client asks months later where an image came from, the log answers in one line. It also catches drift: if partner-model or Midjourney rows start appearing in the "website" column, someone has stepped outside the agreement.

If you want help turning this into a process for your studio, I offer one-to-one AI consultations for small businesses.

What makes an AI image unsafe to hand to a client?

An image is unsafe to hand over if any red flag below applies, so run every image through the list before delivery. Each one is a realistic way a client image goes wrong.

  • A recognisable brand, logo or lettering. Invented or real marks on vehicles, clothing, tools or packaging. Zoom to 100% on edges and backgrounds.
  • A real person's likeness. A face that resembles a public figure or the client's competitor's staff. Regenerate rather than edit.
  • A living artist's name in the prompt. Even where a tool allows it, it undermines any claim that the look is your own. Describe the qualities you want instead.
  • A partner model in a Firefly deliverable. Adobe's commercial-safety statement doesn't extend to it. Check the model name in the generation details, and note that Firefly's AI Assistant chooses between Adobe and partner models unless you name one.
  • A Midjourney image made in a shared channel. Stealth doesn't hide it there. Remake it privately or keep it out of client work.
  • Beta-feature output for an enterprise client who relies on Adobe's indemnity. Adobe's credits FAQ says beta outputs aren't indemnified. Use generally available features for that client.

The same checks protect a yoga studio, a tutoring centre or a removals firm as much as a landscaper. Choose Firefly when the image will carry the client's name in public, use Midjourney where its style range is worth the extra terms and the client has agreed to them, and keep the log either way. If a client needs private Midjourney work as standard, read how Stealth Mode keeps client images private on Pro and Mega before you quote.

Client-work questions about Firefly and Midjourney

Does Adobe indemnify a small studio for Firefly images?

Not on the individual Firefly plans. Adobe's plans page says eligible enterprise customers can receive IP indemnification for select Firefly outputs, and its credits FAQ adds that beta-feature outputs aren't indemnified even for eligible teams and enterprise customers. A small studio relies on Adobe's commercial-use statement and its own checks instead.

Can other people see the images I make for clients in Midjourney?

By default, yes: Midjourney's terms say your content is publicly viewable and remixable. Stealth Mode, available only on the Pro and Mega plans, makes Midjourney use best efforts not to publish what you create, but anything made in a shared Discord channel is visible to that channel either way.

Do I have to tell clients an image was made with AI?

Rules differ by place and platform, so take advice for your situation. As a working practice, say so in your contract and on each delivery note, keep the prompt and tool record, and leave Adobe's Content Credentials attached to Firefly files. Clients then know what they're publishing and can meet any labelling rules on their channels.

Further reads

Show 3 more related tutorials

Sources: Adobe Firefly plans page and FAQ; Adobe generative credits FAQ (updated 16 Nov 2026); Adobe Firefly help pages on style settings and custom models; Adobe partner models page; Adobe Firefly AI Assistant FAQ; Midjourney Terms of Service (effective 27 May 2026); Midjourney docs: Comparing Midjourney Plans, GPU Speed, Style Reference, Version and Upscalers. All checked 17 Nov 2026.

Ready to put AI on the jobs that eat your week?

No data team or big budget needed. In a 1:1 consultation I show small and medium-sized businesses and self-employed people which AI tools they can use, one workflow at a time.

Book a 1:1 call with me