Best European Alternatives to Midjourney
Looking for a European alternative to Midjourney? Midjourney is a US company, offers no meaningful enterprise data protection terms, and still routes much of its workflow through Discord.
European image generation is unusually strong — the team behind Stable Diffusion now builds FLUX in Freiburg, and there are French, Spanish and British options with proper commercial licensing.
How we rank these tools — 4-step process
-
1
European ownership, verified
The company is headquartered and incorporated in the EU, EEA or Switzerland, and processes customer data in Europe. A US parent company disqualifies a tool from this page regardless of where its servers are.
-
2
Category fit and hands-on review
What the tool actually does, who it suits, and where it falls short — checked against the vendor’s own documentation, changelog and pricing page rather than its marketing copy.
-
3
Compliance and pricing check
GDPR posture, hosting location and the prices quoted on this page are verified against the vendor’s public pricing before publication, and re-checked when we revisit the category.
-
4
Position on this page
Placement on this page can be paid, and that can affect which tools appear here and the order they appear in. It never buys a good review: a tool that fails the checks above is not here at any price, and payment does not change the shortcomings we write about. A vendor can ask us to correct a factual error — not to remove a criticism.
Vendors can pay for visibility on this page. It never changes what an entry says about a product, including the criticism, and we earn nothing when you click through to a vendor. Paid placement can affect which tools appear here and the order they appear in. Editorial policy
7 European Alternatives to Midjourney
Black Forest Labs
FLUX models from the team that created Stable Diffusion
Stability AI
Stable Diffusion and open image models you can self-host
Recraft
Generative design with real vector and brand-style control
Freepik AI
Image generation bundled with a large licensed asset library
Photoroom
AI product photography and background generation at scale
Clipdrop
Image editing and generation tools built on open models
Neuroflash
AI image and content generation for marketing teams
Key takeaways
- Article 50 of the AI Act has applied since 2 August 2026: providers must mark synthetic images machine-readably, and deployers must disclose deepfakes — the second duty falls on the business using the tool.
- Midjourney, Inc. is defending four copyright actions, the most recent filed by Warner Bros. Discovery on 4 September 2025.
- Black Forest Labs was founded in Freiburg in 2024 by the people who created Stable Diffusion, so "European models lag behind" is an out-of-date claim.
- Open weights are the only arrangement here where a prompt for an unreleased product never leaves your network.
- Two of the seven options are British rather than EU — adequacy, renewed in December 2025 — and one is a Paris team owned by an American company.
Why people leave Midjourney
Midjourney is very good, and pretending otherwise is the fastest way to lose an argument with a designer. It set the aesthetic standard for this whole category and a large part of what people now consider "AI image quality" is Midjourney's house style.
The reason a European business starts looking is not quality and it is not privacy. It is that since 2 August 2026 the AI Act's Article 50 has applied, and it splits the obligation in two. Providers of systems that generate synthetic images must mark the output in a machine-readable format and make it detectable as artificially generated. Deployers who produce deepfakes must disclose that the content was artificially generated or manipulated. The second of those is you.
Behind that sits a supplier risk that is a matter of public record. Midjourney, Inc. of San Francisco is defending copyright actions brought by artists in January 2023, by more than 4,700 artists in November 2023, by Disney and Universal on 11 June 2025, and by Warner Bros. Discovery on 4 September 2025. That says nothing about who is right. It says a great deal about how stable the ground is under a campaign built on that one model.
- The disclosure obligation lands on you, not on the tool Article 50 puts the marking duty on the provider of the generating system and the disclosure duty on the deployer producing a deepfake. Your agency, your marketing team, your publisher is the deployer. That means the question to ask a supplier is not whether it is compliant, but whether its output carries what you need to demonstrate that you are — machine-readable marking, and provenance you can explain to a client.
- Four copyright actions and counting Artists filed in January 2023, a second action followed in November 2023 covering more than 4,700 artists, Disney and Universal filed on 11 June 2025, and Warner Bros. Discovery on 4 September 2025. None of them is decided. For a brand asset that has to survive five years on packaging, on a vehicle livery or in a television campaign, the relevant question is not whether Midjourney wins but whether you want your asset library to depend on the answer.
- Nothing runs on your own hardware Every Midjourney generation happens on Midjourney's infrastructure, which means every prompt for an unreleased product, an unannounced campaign or a confidential pitch leaves your building. Several models here publish open weights, so generation can happen on a machine with no network connection and there is no record of the request anywhere, because there was no request.
- You cannot fine-tune it on your own material A studio with twenty years of its own photography, a retailer with a catalogue, a publisher with a house illustration style — none of them can teach Midjourney to work in their own idiom. With open weights that is a normal afternoon's work, and the result is a model that produces your aesthetic rather than the aesthetic everyone else's prompts also produce.
What you have to replace, not just match
Work out which of four jobs your Midjourney subscription is doing, because they have four different answers and only one of them is a generative model.
The first is original imagery from nothing — concept art, illustration, hero visuals. The second is assets that have to be edited afterwards: icons, logos, anything that will be recoloured or resized, which a raster image cannot survive. The third is operating on photographs you already own, cutting out products or fixing a shot. The fourth is volume: producing four hundred consistent images rather than one striking one.
Black Forest Labs and Stability answer the first. Recraft answers the second, and is the only one here that does. PhotoRoom and Clipdrop answer the third from opposite directions, one for catalogues and one for repairs. Freepik answers the fourth by combining generation with a stock library, which is the honest admission that generation does not always produce the right thing.
The alternatives compared
| Position | Tool | Headquarters | Pricing | Jurisdiction |
|---|---|---|---|---|
| #1 | Black Forest Labs | Freiburg, Germany | API usage-based / open-weight models free to run yourself | EU (Germany) |
| #2 | Stability AI | London, United Kingdom | Models free for many uses / API usage-based / enterprise licensing above revenue thresholds | United Kingdom (adequacy decision, outside the EEA) |
| #3 | Recraft | London, United Kingdom | Free tier / paid from about $12 per month | United Kingdom (adequacy decision, outside the EEA) |
| #4 | Freepik AI | Malaga, Spain | Free tier with attribution / Premium from about €12 per month | EU (Spain) |
| #5 | PhotoRoom | Paris, France | Free tier / Pro from about €12.99 per month | EU (France) |
| #6 | Clipdrop | Paris, France | Free tier with limits / Pro from about €9 per month / API priced separately | French team, owned by Jasper, a US company |
| #7 | Neuroflash | Hamburg, Germany | Free tier / from €30/month | EU (Germany) |
How each alternative compares to Midjourney
- Which law reaches it. EU (Germany). Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. EU; models can run on your own hardware.
- Source code. Open source, where Midjourney is not: you can read what it does rather than take the description on trust.
- Independently checked. GDPR.
Best for: Teams that want the frontier and want the weights on their own machines
Black Forest Labs GmbH was founded in Freiburg in 2024 by Robin Rombach, Andreas Blattmann and Patrick Esser — the people who built Stable Diffusion — and FLUX became the open-weight benchmark for image generation almost immediately. Anyone still saying European image models are a step behind is describing a situation that ended when that team started a company.
What matters commercially is the licensing and the deployment. Open weights on some FLUX variants mean a model can be downloaded and run on your own hardware: generation with no network connection, no record anywhere of what was asked for, and fine-tuning on your own material.
Set that beside a Midjourney prompt for an unannounced product and the difference is not philosophical. Clear commercial licensing on top answers the question a legal team asks about generated assets, which consumer services tend to leave vague.
The honest limits are those of a young research company. The product is models and an API rather than a finished application with an interface, so there is no equivalent of Midjourney's explore feed or its aesthetic defaults; using it well assumes technical capability; self-hosting needs GPU capacity; and a lab moving this fast is ahead of its own documentation.
What Black Forest Labs does better than Midjourney
- Open weights on some FLUX variants, so generation can happen on a machine with no network connection at all
- Fine-tuning on your own catalogue or house style, which is impossible on Midjourney
- Explicit commercial licensing for generated assets, rather than terms a legal team has to interpret
- German company in Freiburg, so the supplier is established in an EU member state
- FLUX Kontext extends the models into editing existing images rather than only generating new ones
Where Black Forest Labs is a step down from Midjourney
- Models and an API rather than an application, so there is no interface for a non-technical team
- No community feed, no aesthetic presets and none of the discovery that makes Midjourney pleasant
- Self-hosting needs real GPU capacity and someone to maintain the environment
- A young company whose documentation runs behind its releases
Standout against Midjourney. It is the only option here where the people who invented the open image model now sell it from inside the EU, with weights you are allowed to keep.
- Which law reaches it. United Kingdom (adequacy decision, outside the EEA). Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. Your own hardware, or the Stability API.
- Source code. Open source, where Midjourney is not: you can read what it does rather than take the description on trust.
- Independently checked. GDPR.
Best for: Studios and retailers with an archive worth training on
Stability AI Ltd released Stable Diffusion, which is the event that made image generation something anyone could run rather than something you queued for on someone else's service. The tooling, the fine-tunes, the interfaces and most of the research built on top exist because those weights were published.
That remains the reason to use it, and it is the sharpest contrast with Midjourney.
Downloadable weights with community and enterprise licences mean a team can train on its own material and deploy the result on its own hardware — a product photography model built from your catalogue, an illustration model built from your house style — and the images never leave. Its range also extends past images into video and audio, with a hosted API for teams that would rather not run infrastructure.
Two things to check before building on it. Stability AI Ltd is a London company, so this is a United Kingdom adequacy position rather than EU establishment, and running the models locally removes the question entirely. And the models are free for many uses with enterprise licensing applying above revenue thresholds, so establish where your organisation falls before the project rather than after it.
What Stability AI does better than Midjourney
- Downloadable weights with fine-tuning support, so the model can be taught your own material
- By far the largest ecosystem of community tooling around any model on this page
- Video and audio generation alongside images, from the same supplier
- Runs entirely on your own hardware, so confidential prompts never leave the building
- Licences that state what commercial use costs, rather than leaving it to interpretation
Where Stability AI is a step down from Midjourney
- UK company, so adequacy rather than establishment in an EU member state
- Enterprise licensing applies above revenue thresholds, which needs checking before you build
- FLUX leads on the quality of recent work, so this is the ecosystem rather than the frontier
- Fine-tuning needs GPU capacity and expertise that a marketing team will not have in-house
Standout against Midjourney. It is the only option here where a retailer can train a model on twenty years of its own photography and never send an image anywhere.
- Which law reaches it. United Kingdom (adequacy decision, outside the EEA). Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. UK/EU.
- Source code. Closed source, as Midjourney is.
- Independently checked. GDPR.
Best for: Design teams producing icons, logos and anything that will be resized or recoloured
Recraft AI Limited was founded in London in 2022 and generates real vectors, which is the single difference that decides whether what comes out is an asset or a picture. A Midjourney image is finished: scale it and it blurs, change the brand colour and you regenerate and hope the composition survives. What Recraft hands over is a vector file a designer edits directly, with the paths intact.
Custom brand styles are the second idea and they solve the thing that makes generated assets unusable in brand work. Define the style once and every subsequent generation matches it, so thirty icons read as one set rather than as thirty adjacent experiments. Icon and illustration sets, an infinite canvas, mockups and an API sit around it, with a free tier and paid plans from about $12 a month.
It is not the tool for photorealism, where FLUX and Stable Diffusion are far stronger, and it cannot be self-hosted. Being a London company, it is a UK adequacy position rather than EU establishment. And vector output is only worth having if somebody on your team opens vector files — for a marketing department that publishes PNGs, the advantage is theoretical.
What Recraft does better than Midjourney
- Generates genuine editable SVG paths, where Midjourney produces a finished raster image
- Custom brand styles keep a whole set visually consistent instead of individually plausible
- Icon and illustration sets, mockups and an infinite canvas aimed at production rather than exploration
- An API for putting generation inside an existing design workflow
- A free tier, with paid plans from about $12 a month
Where Recraft is a step down from Midjourney
- Not aimed at photorealistic imagery, which is exactly what Midjourney is best at
- A London company, so the position is UK adequacy rather than establishment in a member state
- No self-hosting and no downloadable weights
- Vector output only pays off if a designer is going to open the file
Standout against Midjourney. It is the only tool here that hands you something to edit rather than something to accept, which is the difference between an asset and a screenshot.
- Which law reaches it. EU (Spain). Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. EU (Spain).
- Source code. Closed source, as Midjourney is.
- Independently checked. GDPR.
Best for: Marketing teams on a deadline who need something usable rather than something novel
Freepik Company S.L. has operated from Málaga since 2010, which makes it one of the more settled companies on this page rather than a recent arrival. It pairs AI image and video generation with a very large library of licensed stock assets, vectors and templates under one subscription, with a free tier that requires attribution and Premium from about €12 a month.
The argument against Midjourney is practical rather than technical. Generation does not always produce the right thing, and when it does not, Midjourney's answer is another prompt. Freepik's answer is something already drawn, one click away, under the same licence and the same invoice — which on a Thursday afternoon with a campaign going out on Friday is the difference between shipping and not.
It runs multiple underlying models rather than one, so the output is not tied to a single lab's aesthetic or roadmap, and it adds the operational work around images: background removal, upscaling, and an API. The trade is depth. It is a platform assembling capability rather than a lab producing it, so on pure generation quality the model labs lead, and there is no self-hosting.
What Freepik AI does better than Midjourney
- A licensed stock library beside the generator, so a failed prompt is not a dead end
- One subscription and one licence covering both generated and pre-drawn assets
- Several underlying models rather than one lab's house style and one lab's roadmap
- Spanish company operating since 2010, with EU processing and a long track record
- Background removal, upscaling and an API included rather than assembled from other tools
Where Freepik AI is a step down from Midjourney
- Generation quality trails the model labs, because it assembles capability rather than producing it
- No self-hosting and no downloadable weights
- The free tier requires attribution, which rules it out for most commercial work
- Breadth rather than depth: nothing here is the best version of itself
Standout against Midjourney. It is the only option on this page where the answer to "the prompt is not working" is an asset somebody already drew, under the licence you are already paying for.
- Which law reaches it. EU (France). Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. EU (France).
- Source code. Closed source, as Midjourney is.
- Independently checked. GDPR.
Best for: Sellers and brands processing a product catalogue
PhotoRoom operates from Paris and applies AI to the tedious half of image work rather than the interesting half. Cutting a product out of its background, placing it on white or into a generated scene, and doing it identically across a catalogue is the job standing between a seller and a listing that converts, and it is the job nobody has time for.
That is a different problem from the one Midjourney solves, and the distinction matters when a team reaches for a generator by reflex. Midjourney can produce a beautiful image of a product like yours; it cannot process the photograph of the product you actually sell.
Batch editing turns that from a feature into an argument — one photograph is two minutes and four hundred is a week — and the API extends the same processing into an e-commerce pipeline so it happens on upload rather than as a task somebody remembers.
The scope is deliberately narrow: commercial product imaging, not a general editor and not a generative model. There is a free tier with Pro from about €12.99 a month, batch editing needs the paid tier, and complex retouching still needs a real editor. Product photographs are frequently unreleased when they are processed, which is a good reason to care that the processing is in France.
What PhotoRoom does better than Midjourney
- Processes photographs you already own, which no generative model can do for you
- Batch editing across a whole catalogue in one operation, with consistent framing and background
- An API that puts the same processing inside an e-commerce upload pipeline
- French company with EU processing, for product images that are usually unreleased at the time
- A free tier with Pro from about €12.99 a month
Where PhotoRoom is a step down from Midjourney
- Not a generative model at all, so it replaces none of what Midjourney is used for creatively
- Narrow by design: product imaging, not general editing or illustration
- Batch editing, the feature that makes the case, requires the paid tier
- No self-hosting and no way to run it on your own infrastructure
Standout against Midjourney. It is the only tool here aimed at the four hundredth photograph rather than the first one, which is where a catalogue actually gets stuck.
- Which law reaches it. French team, owned by Jasper, a US company. Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. EU (France).
- Source code. Closed source, as Midjourney is.
- Independently checked. GDPR.
Best for: Photographers and marketers fixing shots that already exist
Clipdrop is a set of focused operations rather than a generative playground, and the list is the product: cleanup to remove an object convincingly, uncrop to extend an image past its frame, relight to change how a photograph was lit after the fact, background removal and upscaling. Each does one thing a photographer or marketer actually needs done, built on open generation models underneath.
That makes it complementary to Midjourney rather than competitive with it, and worth having for a reason the AI Act sharpens. Article 50 exempts assistive editing functions that do not substantially alter the input data, so an operation that repairs a photograph you took sits in a different place from one that generates a scene that never existed. Batch processing, an API and desktop and mobile applications cover the different ways that work arrives.
The ownership needs stating plainly rather than buried. The team is French and based in Paris with EU processing under the GDPR, and the company is owned by Jasper, an American business. European processing, American ownership: the two halves point in different directions, and only your own procurement rule decides which half it was written about. There is a free tier with Pro from about €9 a month and the API priced separately.
What Clipdrop does better than Midjourney
- Relights, uncrops and cleans up photographs you already have, which no amount of prompting replaces
- Batch processing and an API for volume work rather than one image at a time
- Desktop and mobile applications alongside the web tools
- French team with EU processing under the GDPR
- Free tier with Pro from about €9 a month
Where Clipdrop is a step down from Midjourney
- Owned by Jasper, an American company, so European ownership is not what you are buying
- Not a generative model in its own right, so it answers none of the creative brief
- No self-hosting and no downloadable weights
- Free tier limits bite quickly on any real volume of work
Standout against Midjourney. It is the only tool here that can change how a photograph was lit after it was taken, which is the one operation prompting cannot reach.
- Which law reaches it. EU (Germany). Midjourney is run from the United States, so the CLOUD Act obliges the provider to hand over data on a valid order regardless of which country the servers are in.
- Where the data sits. EU (Germany).
- Source code. Closed source, as Midjourney is.
- Independently checked. GDPR.
Best for: German-language marketing teams who need the copy more than the visual
Neuroflash GmbH is in Hamburg and is a copy generation platform with image generation attached, which is why it sits last on a page about image tools. It is here because a large share of the teams paying for Midjourney are marketing departments, and for a marketing department the scarce thing is rarely the hero image — it is thirty product descriptions, four ad variants and a newsletter, all in German, all this week.
Its Brand Hub stores brand voices that shape tone and vocabulary across everything generated, which is the difference between copy a marketer ships and copy that reads like every other company using the same tool. Around it sit ChatFlash for conversational refinement, an SEO editor scoring against target keywords, plagiarism detection, a performance predictor and accompanying image generation. German output is markedly stronger than American tools translating into German, and it also handles English, French, Spanish and Italian.
Nobody should choose it as an image tool. The generation is a supporting feature, not a frontier model, and if you need the hero visual the six entries above this one are where to look. Plans start from €30 a month after a free tier, which is a real step up, and generated copy still needs editing before it publishes.
What Neuroflash does better than Midjourney
- Generates the campaign copy as well as the image, which is usually the actual bottleneck
- Brand voices stored once and applied to everything generated, so output stays in one register
- German output written for German rather than translated into it, from a Hamburg company
- SEO scoring, plagiarism detection and performance prediction built in rather than bought separately
- EU processing, so campaign plans and unreleased product details stay in European jurisdiction
Where Neuroflash is a step down from Midjourney
- The image generation is a supporting feature, nowhere near Midjourney or FLUX in quality
- From €30 a month, a real step up from the free tier and from most of this page
- Strongest in German; the other languages are good rather than exceptional
- Wrong tool entirely if what you need is a single striking visual
Standout against Midjourney. It is the only entry here that writes the words next to the picture, which for a marketing team is the half of the brief that actually runs late.
What actually breaks when you switch
The house style goes with the tool. Midjourney has a recognisable look, and a brand that has been generating with it for a year has an asset library in that look. Swapping models mid-campaign produces a visible seam, so plan the switch at a campaign boundary and regenerate the recurring assets deliberately rather than discovering the mismatch in a shop window.
Prompts do not port. A prompt is tuned to one model's interpretation, and the same words produce something else entirely on FLUX or Stable Diffusion. Budget a week of relearning for whoever writes them, and resist the urge to grade the new tool against prompts written for the old one — that comparison is guaranteed to be unfair and is the usual reason a trial fails.
And write down what your supplier puts in the file. With Article 50 applying since 2 August 2026, the marking on generated output is now part of what you are buying, and the deployer disclosure is yours to make.
Providers whose systems were already on the market before that date have until 2 December 2026 for the marking duty, so ask for the date as well as the answer — "we will check later" is a poor answer to give a client whose campaign is already live.
What does the AI Act actually require of us when we generate an image?
Article 50, which has applied since 2 August 2026, separates two roles. The provider of the system that generates synthetic image content must ensure the output is marked in a machine-readable format and detectable as artificially generated or manipulated, using solutions that are effective, interoperable, robust and reliable as far as is technically feasible. That obligation belongs to the model supplier.
The deployer — the organisation using the system — must disclose when content is a deepfake, with a lighter requirement for artistic and creative work, where it is enough to note that manipulated content exists without spoiling the work. There are carve-outs for assistive editing that does not substantially alter the input, and for law-enforcement uses.
The practical consequence for a buyer is a procurement question rather than a legal one: can this supplier tell you what marking its output carries, and will it put that in writing. A supplier that cannot answer is one whose output you are attaching your own disclosure obligation to blind.
Do the lawsuits against Midjourney actually affect us?
Not directly, and it would be dishonest to suggest a European marketing team is about to be sued by Disney. The suits are against Midjourney, Inc. and concern how the models were trained.
What they affect is the durability of what you produce. Four actions are outstanding — artists in January 2023, more than 4,700 artists in November 2023, Disney and Universal on 11 June 2025, Warner Bros. Discovery on 4 September 2025 — and any of them could change what the product may generate, what it may keep generating, or what it costs. An asset library built on one supplier inherits that uncertainty.
The mitigation is not moral, it is architectural. Open weights you hold a copy of cannot be changed by someone else's judgment, and a supplier with explicit commercial licensing gives your legal team a document rather than a hope.
Is running an image model ourselves realistic?
More realistic than it was, and still not free. Open-weight FLUX variants and Stable Diffusion models can be downloaded and run on your own hardware, which means generation with no network connection, no record of what was asked for, and the ability to fine-tune on material you own.
The costs are a GPU and a person. A machine capable of running a current image model at usable speed is a real purchase or a real hourly rate, and somebody has to keep the environment working. For a small marketing team that is out of proportion; for an agency generating at volume, or a studio with confidentiality obligations to clients, it usually is not.
The middle path most organisations take is an API from the same lab, on the understanding that the weights exist and can be brought in-house if the relationship or the regulation changes. That optionality is worth something on its own.
How European is this list, honestly?
Four of the seven are unambiguously EU: Black Forest Labs in Freiburg, Freepik in Málaga, PhotoRoom in Paris and Neuroflash in Hamburg.
Stability AI Ltd and Recraft AI Limited are both London companies, which means a United Kingdom adequacy position — renewed by the Commission in December 2025 — rather than establishment in an EU member state. For most buyers that is a distinction without a difference; for a public-sector procurement rule that names member states, it is the whole answer.
Clipdrop needs stating plainly: the team is in Paris and processing is in the EU, and the company is owned by Jasper, an American business. So it is EU processing without EU ownership, and whether that satisfies you depends entirely on which of the two your rule was written to catch.
Which one to pick
If you want the frontier and you want it inside the EU, Black Forest Labs is the answer, and the open weights are the part that matters: a prompt for an unreleased product never has to leave the building.
If your advantage is your own archive, Stability AI is the only route here to a model trained on it. Check where your organisation falls against the revenue thresholds in its licensing before the project starts, and note that London is adequacy rather than EU establishment.
If the output has to be edited by a designer rather than accepted by a marketer, Recraft is the only tool on this page that produces something editable, and nothing else comes close.
And if you are a marketing team on a deadline, be honest about which problem you have. Freepik when the prompt fails, PhotoRoom when it is four hundred product shots, Clipdrop when the photograph exists and is merely wrong — and Neuroflash when the thing actually running late is the copy.
Frequently Asked Questions
Black Forest Labs if you want state-of-the-art generation with weights you can run yourself. Stability AI if fine-tuning on your own catalogue is the point. Recraft if the output has to be editable rather than merely finished. Freepik if you need a fallback when the prompt does not work. PhotoRoom for product catalogues, Clipdrop for repairing photographs you already have.
Article 50 has applied since 2 August 2026. The provider of the generating system must mark output in a machine-readable format and make it detectable as artificially generated or manipulated. The deployer — your organisation — must disclose deepfake content, with a lighter requirement for artistic and creative work.
One timing detail matters when you ask a supplier: providers of systems already on the market before 2 August 2026 have until 2 December 2026 to comply with the marking duty, so an answer of "not yet" is not necessarily an answer of "never". Ask what the output carries, and ask when.
Yes, in four actions. Artists Sarah Andersen, Kelly McKernan and Karla Ortiz filed on 13 January 2023 over training on images scraped from the web; a second action in November 2023 concerned the work of more than 4,700 artists; Universal Pictures and Disney filed on 11 June 2025; and Warner Bros. Discovery filed on 4 September 2025. None is resolved, and none of them is a finding against Midjourney.
Yes, with open weights. Some FLUX variants from Black Forest Labs and the Stable Diffusion family from Stability AI can be downloaded and run locally, which means no prompt leaves your network and no record of a request exists anywhere. It needs GPU capacity and somebody to maintain the environment, which is the real cost and the reason most teams use a hosted API from the same labs instead.
No. It launched a web application in August 2024 alongside version 6.1, so the Discord bot is one route rather than the only one. If you passed on Midjourney because your organisation would not adopt Discord, that objection has expired and you should evaluate it on its merits rather than on its interface.
Recraft, and only Recraft. The output is a genuine vector file your designer opens and edits in the same way as anything drawn by hand, with custom brand styles so a set of thirty icons holds together. Everything else here, Midjourney included, produces raster images: scale them and they blur, change the brand colour and you regenerate and hope.
Not with Midjourney. With Stability AI's downloadable weights you can, and it is the main reason to use them: a product photography model trained on your catalogue, or an illustration model trained on your house style, running where the images never leave. Black Forest Labs supports fine-tuning on its open-weight variants as well. It needs GPU capacity and expertise, so treat it as a project rather than a setting.
It differs by supplier and this is where reading the licence pays. Black Forest Labs publishes clear commercial licensing for generated assets, Stability applies community and enterprise licences with thresholds tied to revenue, and Freepik bundles generation with a licensed stock library so the fallback asset comes with its own terms. Whatever you choose, the document you want is the one your legal team can read without asking anybody a question.
Usually irrelevant, because they are priced differently rather than lower. Open weights are free to run and cost GPU time; Recraft and Freepik have free tiers and paid plans from roughly $12 and €12 a month; PhotoRoom Pro is about €12.99 and Clipdrop Pro about €9. The decision here is provenance, editability and where generation happens, and picking on headline price is how teams end up with the wrong tool cheaply.
Who worked on this review
Explore More European Alternatives
Discover privacy-focused European alternatives to other popular US tech services.