There is a question that decides whether you can use a generated image, and it is not "does it look good." It is "if a rights holder comes after this, who stands between me and the lawsuit?" That word is indemnification, and in 2026 the answer differs sharply by tool. Adobe Firefly puts a commercial indemnity behind its paid output. Midjourney does not, and Midjourney is currently the named defendant in a consolidated lawsuit brought by Disney, Universal, and Warner Bros. Discovery. Stable Diffusion and FLUX are open-weights tools that indemnify nobody. This lesson reads the actual terms each tool offers, explains what indemnification does and does not cover, and hands you a "Can I ship this for a client?" decision tree you can run before a single asset reaches a deliverable.

Why Indemnification Is the Question, Not Image Quality

Picture the brand and visual designer on a Wednesday, generating fifteen social variants for a launch. The images look great from four different tools. The actual decision is not which looks best; it is which one marketing legal will sign off on, because the company, not the model, ships the asset, and the company carries the liability if that asset turns out to infringe someone's copyright or trademark. Generative image models are trained on enormous corpora scraped from the open internet, and that training data includes copyrighted work the model can, under the right prompt, reproduce closely enough to create real legal exposure.

Indemnification is a contractual promise from the vendor that says, in effect, "if you use our output as permitted and someone sues you for IP infringement over it, we will defend you and cover the costs." It does not make infringement legal. It does not stop a lawsuit from being filed. What it does is shift the financial risk from you to the vendor, within limits and conditions. For a working designer that distinction is everything, because it converts an unbounded, terrifying, company-killing risk into a known, bounded one that a legal team can actually evaluate and accept. The image quality question is real but secondary. The indemnification question is the one that gates whether the asset can ship at all.

Adobe Firefly: The Indemnified Option

Adobe Firefly is, as of 2026, the major image model that carries commercial indemnification on its paid Creative Cloud plans. Adobe's pitch has always been that Firefly was trained primarily on Adobe Stock, openly licensed content, and public-domain material, which is the legal foundation that lets Adobe put its own money behind the output. When you generate with Firefly on a qualifying paid plan and use the result within the terms, Adobe offers to indemnify your business against third-party IP claims arising from that output.

For the brand designer staring down a launch, this is why Firefly is the default first reach. It is the tool where, when marketing legal asks "who covers us if this gets challenged," there is an actual answer with a company's name on it. That answer has limits - the indemnity applies to qualifying output, used as permitted, and does not cover everything a creative might do downstream - but it exists, and "an indemnity exists, here are its boundaries" is a fundamentally different legal posture than "no indemnity, you are on your own."

Indemnification does not mean the output cannot infringe. It means that if it does, you are not the only one holding the bill. For a company shipping client work, that is the line between a usable tool and an unusable one.

The Firefly Caveat Worth Knowing

Firefly is not a free pass, and a careful designer knows the asterisk. There is a 2024-2025 disclosure, covered in the next lesson, that roughly 5 percent of Firefly's training set included AI-generated images, some sourced from Midjourney, which complicates the clean "ethically trained on licensed data only" story. The indemnity still stands as a contractual matter, but the provenance story is messier than the marketing implies, and a designer who quotes "Firefly is completely ethically trained" without the asterisk will get corrected by a legal team that has read the disclosure. Use Firefly for the indemnity; do not oversell its purity.

Midjourney: No Indemnity, and an Active Lawsuit

Midjourney sits at the opposite end. It offers no commercial indemnification. If you generate a Midjourney image, ship it for a client, and a rights holder sues, Midjourney's terms do not promise to defend or cover you. The risk lands on you and the company that shipped the asset. That alone would make Midjourney a cautious choice for client work; the litigation makes it a documented one.

The case material is specific and worth knowing precisely, because vague "AI is being sued" hand-waving is exactly what makes designers sound unserious in a legal conversation. Disney and Universal filed suit against Midjourney in June 2025, alleging the tool reproduces their copyrighted characters. Warner Bros. Discovery joined with its own suit in September 2025. The cases were consolidated. As of May 2026, the parties are in private mediation with a court-ordered alternative dispute resolution deadline of August 19, 2026 - which means the matter is in mediation, not active discovery, and has not been decided. Nobody knows the outcome yet, and a responsible designer does not pretend to.

What the Lawsuit Actually Means for Your Work

The practical reading is not "Midjourney is illegal." It is "Midjourney is the tool with the highest documented IP risk and no indemnity to offset it, while a major studio lawsuit over exactly the kind of reproduction designers worry about is live and unresolved." That combination is why Midjourney is hard to defend for client deliverables, especially anything touching recognizable characters, brands, or styles that a studio might claim. For personal exploration, moodboards, and internal ideation that never ships, the calculus is different and far more permissive. The risk is about what you ship and to whom, not about whether you may open the tool.

FLUX and Stable Diffusion: Open Weights, No Cover

FLUX and Stable Diffusion are open-weights models, which is a genuine strength for control, local running, and customization, and a genuine gap for legal cover. Because they are open-weights tools you can run yourself, there is no vendor sitting behind your output offering to defend you. The flexibility that makes them powerful for technically minded designers is the same property that leaves you fully exposed: you are the operator, the corpus provenance is often opaque, and there is no indemnity because there is no central commercial provider promising one for your specific use.

This does not make them unusable. It makes them tools whose IP risk you carry entirely yourself, which is acceptable for some contexts and not others. A graphic designer experimenting with FLUX for a concept that will be redrawn by hand before it ships is in a very different position than one who pastes a raw FLUX output into a client's paid campaign. Recraft, by contrast, is worth checking on its own current terms rather than assuming; its positioning around vector and brand work means its commercial terms are a thing you read for the specific plan you are on, not a thing you guess. The general rule for all of these: open-weights and most non-Adobe providers leave the IP risk with you, so confirm the terms for your exact plan before client output, never after.

What Indemnification Does Not Cover

It is just as important to know the limits, because a designer who treats an indemnity as a magic shield will misuse it and lose it. Indemnities are conditional. They typically apply only to the vendor's qualifying output used as permitted under the terms; they often exclude cases where you prompted the model to reproduce a specific known work, brand, or character; and they do not cover trademark or right-of-publicity problems you create downstream, like slapping a generated image next to a competitor's logo or generating a celebrity's likeness. The indemnity covers the model's contribution to infringement, broadly, not your decision to deliberately steer it toward someone else's protected property.

This is why prompting hygiene matters even with the indemnified tool. "A friendly cartoon mouse in red shorts" is asking for a lawsuit no indemnity will save you from, because you steered it. "A friendly cartoon woodland creature in our brand palette" is a different posture entirely. The indemnity protects you when you use the tool as a tool; it abandons you when you use it as a copy machine for known IP. Read the terms, respect the exclusions, and never let "we are indemnified" become an excuse to prompt for protected work.

The "Can I Ship This for a Client?" Decision Tree

Here is the artifact this lesson exists to give you. Run it before any generated image reaches a client deliverable. It is built so a designer and a legal reviewer can run it together, because the answer has to satisfy both.

1. Will this ship externally, or is it internal only? If it is moodboard, ideation, or internal exploration that never reaches a client or the public, IP risk is low and most tools are fine. If it ships externally, continue.
2. Did the prompt steer toward known IP? If you prompted for a recognizable character, brand, celebrity, or distinctive protected style, stop. No indemnity covers deliberate reproduction; redraw or re-prompt away from the protected property.
3. Does the tool carry commercial indemnification for your plan? Firefly on a qualifying paid Creative Cloud plan does. Midjourney, Stable Diffusion, and FLUX do not. Recraft and others: read your specific plan's terms.
4. If no indemnity, can the asset be made safe another way? Use it only as a concept to be redrawn or replaced by licensed or original work before shipping, or move the generation to the indemnified tool.
5. Has legal seen the posture for this campaign? For anything high-visibility or brand-critical, the indemnity status and prompt history are a one-line note legal can sign off on. "Generated in Firefly on a paid plan, no IP-targeting prompts" is a sentence that ends an argument.

The tree resolves to a clean verdict: ship, ship-after-redraw, or do-not-ship. It turns a vague anxiety into a documented decision, which is exactly what a designer needs when a CPO or a legal team asks why a particular asset was cleared.

A Worked Example: Clearing a Launch Asset

The brand designer needs a hero image of a "joyful family celebrating a milestone" for a product launch. Run the tree. Ships externally? Yes, it is the launch hero. Continue. Steered toward known IP? No - "joyful family" is generic, no character, brand, or celebrity. Good. Indemnified tool? They generate in Firefly on a paid Creative Cloud plan, so yes. Made safe another way needed? Not required, since the tool is indemnified and the prompt is clean. Legal seen the posture? They attach the one-line note: "Hero generated in Adobe Firefly, paid CC plan, generic prompt, no IP targeting." Legal signs off in minutes.

Now run the same brief through Midjourney. Ships externally? Yes. Steered toward IP? No. Indemnified? No, and there is live studio litigation. The tree now resolves to ship-after-redraw or do-not-ship: the Midjourney image can serve as a composition reference, but the shipped asset should come from the indemnified tool or be replaced by licensed or original work. Same image quality, completely different shipping decision, and the difference is entirely the indemnity posture, not the pixels.

The Four Indemnity Postures, on One Map

It helps to collapse the whole landscape into four postures, because the specific tool names will change but the postures are stable. Posture one is indemnified-and-conditional: a vendor stands behind qualifying output used as permitted. Adobe Firefly on a paid Creative Cloud plan is the canonical 2026 example. This is the posture you want for client work, with full awareness of the conditions that bound it. Posture two is no-indemnity-plus-active-litigation: the tool offers no cover and is itself a named defendant in unresolved IP cases. Midjourney sits here, with the consolidated Disney, Universal, and Warner Bros. Discovery suit in mediation toward the August 19, 2026 ADR deadline. This is the highest-risk posture for anything that ships externally.

Posture three is open-weights-no-cover: powerful, controllable, locally runnable, and entirely your liability. FLUX and Stable Diffusion live here. The flexibility is real and so is the exposure; there is no central provider to defend you because you are the provider. Posture four is read-the-plan: a tool whose terms vary enough by plan and use that the only honest answer is "check the specific terms you are on." Recraft and many others belong here, and the mistake is guessing instead of reading. When a new tool launches next quarter, you will not need a new lesson. You will need to ask which of these four postures it occupies, and the decision tree will resolve the same way.

The value of compressing it to four postures is that it survives the churn of the tool market. Models come and go, terms get rewritten, lawsuits settle or escalate, but a designer who thinks in postures rather than brand names can place any new tool on the map in a single question and reach a defensible shipping decision without waiting for someone else to tell them whether the tool is "safe."

A Common Mistake: Treating the Indemnity as a Permission Slip

The single most expensive misreading of this entire topic is the one that sounds the most reasonable: "We use the indemnified tool, so we are covered, so we can generate whatever we need." This collapses the moment it meets a real exclusion, and it collapses in the worst possible way - after the asset has shipped. The indemnity is not a permission slip to generate anything; it is a conditional backstop for output produced and used as the terms permit. The designer who hears "indemnified" and stops reading is the one who prompts the indemnified tool for "a beloved animated character in our brand colors," ships it, and discovers that deliberate reproduction of known IP is exactly the kind of thing the indemnity excludes.

Walk the failure concretely. A brand designer needs a playful mascot for a campaign and prompts Firefly toward something that lands a little too close to a recognizable studio character because the prompt named the vibe by referencing it. The output is generated in the indemnified tool, so the team feels safe and ships. Then a rights holder notices. The team reaches for the indemnity and finds the exclusion: prompts that steer toward known IP are not covered, because the infringement came from the operator's deliberate choice, not the model's general behavior. The indemnified tool did not save them, because they used it as a copy machine for protected work. The lesson is permanent: the indemnity covers the tool used as a tool, never the tool used to launder someone else's IP, and "we are indemnified" is the beginning of the analysis, not the end of it.

The fix is to keep prompt hygiene and indemnity status as two separate gates that both must pass. Gate one: did the prompt steer toward known IP? If yes, no tool and no indemnity saves you - redraw or re-prompt away. Gate two: does the tool carry indemnity for this plan and use? Only an asset that passes both gates is clear to ship. Designers who fuse the two gates into "are we on the indemnified tool" ship the exact assets that get the company sued, with a false sense of safety that makes the mistake more likely, not less.

How This Changes Your Week-to-Week Practice

None of this matters if it lives only in a lesson, so make it operational. On Monday, set your default generation tool for any client-facing imagery to the indemnified option, so the safe path is the path of least resistance and an exposed tool requires a deliberate exception rather than a default. On Tuesday, when you reach for a non-indemnified tool because it is genuinely better for a concept, write one line that records why and confirms the output will be redrawn or replaced before it ships - the start of the provenance discipline two lessons from now. On Wednesday, when marketing asks for fifteen variants under deadline, you already know which tool clears legal fastest, so the six-hour scramble becomes a ninety-minute run because the indemnity question was answered before the deadline, not during it.

The thread through the whole week is that the indemnity decision is cheap when made early and ruinous when made late. A designer who resolves the posture before generating spends seconds; a designer who generates first and asks later risks a do-over at best and a legal exposure at worst. Build the decision tree into the front of your workflow, not the back. The asset that was cleared before it was made is the asset that ships on time, and the designer who can state its posture in one sentence is the designer legal stops second-guessing.

How to Talk About This Without Overclaiming

The fastest way to lose a legal conversation is to overclaim. Do not say "Firefly is legally safe" - say "Firefly carries commercial indemnification on paid plans, which shifts IP risk to Adobe within their terms." Do not say "Midjourney is illegal" - say "Midjourney offers no indemnity and is in unresolved litigation with major studios, so it is high-risk for external client work." Do not say "open-source models are fine" - say "FLUX and Stable Diffusion are open-weights with no indemnity, so the IP risk sits entirely with us." Precision is credibility. A designer who can state the actual posture of each tool, with the actual dates and the actual conditions, is a designer legal will trust to make the call. A designer who hand-waves "AI and copyright, it's complicated" is a designer who will have the decision taken away from them.

Key Takeaways

• The gating question for any generated image is not quality but indemnification: if a rights holder sues, who stands between you and the cost. Indemnification shifts that risk from you to the vendor, within limits; it does not make infringement legal.
• Adobe Firefly carries commercial indemnification on paid Creative Cloud plans, built on its mostly-licensed training claim. It is the default reach for client work because legal gets an actual answer to "who covers us."
• Midjourney offers no indemnification and is the named defendant in a consolidated lawsuit: Disney and Universal filed June 2025, Warner Bros. Discovery joined September 2025, and as of May 2026 the parties are in private mediation with a court-ordered ADR deadline of August 19, 2026 - not yet decided.
• FLUX and Stable Diffusion are open-weights with no indemnity, so the IP risk sits entirely with you. Recraft and other providers: read your specific plan's terms rather than assuming.
• Indemnities are conditional: they cover qualifying output used as permitted, exclude prompts that deliberately target known IP, and do not cover downstream trademark or right-of-publicity problems. Prompting hygiene matters even with an indemnified tool.
• Run the "Can I ship this for a client?" decision tree - external? IP-targeting prompt? indemnified tool? safe another way? legal seen it? - to convert IP anxiety into a documented ship, ship-after-redraw, or do-not-ship verdict.
• Never overclaim. State the actual posture with actual dates and conditions; precision is what makes legal trust you to make the call.