Every European AI image & video API reviewed
Freiburg, Germany
Founded 2024
API usage-based / open-weight models free to run yourself
Best for: Teams that want frontier generation quality with an open-weight exit path
Black Forest Labs is the German lab founded in 2024 by the creators of Stable Diffusion, and the FLUX family it produces is the strongest image model line built in Europe. With FLUX 3, released in July 2026, image and video generation sit behind the same multimodal API, generating video at 1080p.
Black Forest Labs operates from Freiburg under German jurisdiction, and the licensing is deliberately layered: FLUX Pro is closed and API-only, FLUX Dev is open-weight for non-commercial use at near-Pro quality, and FLUX Schnell is openly licensed for fast generation. That structure is why so many other products in this list — including Nebius AI Studio — serve FLUX models rather than models of their own.
For a buyer the practical consequence is leverage. You can build on the API for speed and keep the open weights as an exit, running them on your own hardware or a European GPU cloud if pricing or terms change. Check which variant you are calling, because the licences differ.
What Black Forest Labs does well
- Frontier image quality from a European lab
- FLUX 3 puts image and video on one API
- Open weights for the Dev and Schnell variants
- German jurisdiction, GDPR applies directly
- Usage-based API pricing, no seats
- Models can be run on your own hardware
Where Black Forest Labs falls short
- FLUX Pro is closed and API-only
- Licence differs per variant — read before shipping
- Young company compared with the US incumbents
- Self-hosting the open weights is a real operational commitment
Standout feature. The only European lab producing frontier image models with open weights and video on the same API.
London, United Kingdom
Founded 2019
Models free for many uses / API usage-based / enterprise licensing above revenue thresholds
Best for: Teams that want to fine-tune image models on their own data
Stability AI is the London company behind Stable Diffusion, and it remains the most open of the model builders here: it has published downloadable weights since 2022 and still releases open models alongside its REST API for image and video generation.
Stability AI was founded in 2019 and is UK-based, which puts it under the EU adequacy decision rather than inside the EEA — transfers are lawful without additional safeguards, a distinction that matters to a small number of organisations with intra-EEA processing requirements and to almost nobody else. Its commercial model is unusual: models are free for many uses, the API is usage-based, and enterprise licensing applies above revenue thresholds.
What Stability offers that the others do not is the fine-tuning ecosystem. Because the Stable Diffusion family has been open longest, the tooling, LoRAs and training recipes around it are deeper than for any other model line, which is decisive if your product needs a model trained on your own catalogue rather than a general one.
What Stability AI does well
- Open model weights you can download and fine-tune
- Deepest fine-tuning and LoRA ecosystem
- Image and video generation on one REST API
- Free for many uses, usage-based above that
- UK adequacy decision covers EU transfers
Where Stability AI falls short
- UK rather than intra-EEA processing
- Enterprise licensing kicks in above revenue thresholds
- Frontier quality now contested by FLUX
- Fine-tuning is real work, not a setting
Standout feature. The open-weight ecosystem: no other provider here has as much community tooling for training on your own data.
Amsterdam, Netherlands
Pay per generation
Best for: Teams that want the published open weights without operating GPUs
Nebius AI Studio serves the leading open image models — FLUX Schnell, FLUX Dev and SDXL — through a single hosted API, so a product can run the weights the labs publish without buying or operating GPUs.
Nebius AI Studio is the inference arm of Nebius Group N.V., the Amsterdam-headquartered AI infrastructure company listed on Nasdaq. That matters more than it sounds: Nebius builds and runs its own European datacentres, which is a different proposition from the US aggregators that broker capacity they do not own. For a team that wants open models served from Europe, the compute underneath is European too.
Because everything in the image catalogue is open-weight, there is no lock-in to escape. Moving from Nebius to your own GPUs, or to another host of the same models, is a change of endpoint. The API surface is OpenAI-compatible, and a playground lets you compare models before wiring anything up.
What Nebius AI Studio does well
- Hosts FLUX and SDXL — the models the labs publish
- Own European datacentres rather than brokered capacity
- OpenAI-compatible API surface
- Pay per generation, no subscription
- Same account and key as its text and embedding models
- No lock-in: every image model is open-weight
Where Nebius AI Studio falls short
- A host rather than a model builder — no models of its own
- No specialised generation like vectors or avatars
- Nasdaq-listed parent with R&D outside Europe too
- Image catalogue narrower than a general aggregator
Standout feature. Open-weight image models served from datacentres the company builds and runs itself, inside the EU.
London, United Kingdom
Founded 2022
Free tier / paid from about $12/month
Best for: Design systems that need real vectors rather than traced bitmaps
Recraft is the only API in this category that generates native SVG: genuine vector geometry with clean paths you can edit node by node, rather than a raster image traced into approximate curves. For icon sets, illustrations and anything that has to scale or be recoloured inside a design system, that is the entire decision.
Recraft was founded in 2022 by Anna Veronika Dorogush, creator of the CatBoost machine-learning framework, and operates from London — UK jurisdiction under the EU adequacy decision. The V4.1 models released in May 2026 continue a line aimed at professional design work rather than general-purpose image generation.
Beyond vectors, the API covers raster generation, inpainting, outpainting, background removal, batch jobs and asynchronous processing, with brand style controls that keep output consistent across a whole set. Pricing starts with a free tier and paid plans from around $12 a month, with API usage billed separately.
What Recraft does well
- Native SVG output, unique in this category
- Brand style controls for consistent sets
- Inpainting, outpainting and background removal
- Batch and asynchronous jobs for volume
- Free tier for evaluation
Where Recraft falls short
- UK rather than intra-EEA processing
- No open weights or self-hosting path
- No video generation
- Aimed at design work rather than photorealism
Standout feature. Genuine vector output — no other generation API here produces editable SVG geometry.
London, United Kingdom
Founded 2017
From $29/month (Starter)
Best for: Product and training teams generating presenter video at volume
Synthesia generates presenter and avatar video from a script rather than from a prompt, which makes it a different kind of tool from everything else on this list — closer to a video production pipeline than to a generative model endpoint.
Synthesia was founded in 2017 and operates from London under the UK adequacy decision. It supports well over a hundred languages, which is the feature that decides most purchases: producing the same training or product video in twenty languages is a translation and re-record project with human presenters, and a re-render with Synthesia.
The API matters for teams that generate video programmatically — course content per product SKU, personalised onboarding, localised release notes — rather than for one-off marketing pieces made in the editor. Pricing starts from $29 a month on the Starter plan, with API access on higher tiers.
What Synthesia does well
- Script-to-video with AI presenters
- 120+ languages from one script
- API for programmatic video generation
- Templates and brand kits for consistency
- Established company with enterprise customers
Where Synthesia falls short
- UK rather than intra-EEA processing
- Avatar video only — not general video generation
- Subscription-based rather than pure usage billing
- Avatars remain recognisably synthetic to many viewers
Standout feature. The only provider here that turns a script into presenter video in over a hundred languages.
Paris, France
Founded 2019
Free tier / Pro from €12.99/month
Best for: Marketplaces and sellers processing product photography at volume
PhotoRoom solves the problem most companies actually have, which is not a blank canvas but a catalogue of inconsistent supplier photography. Its API removes and replaces backgrounds, generates studio-style product shots and processes images in volume.
PhotoRoom was founded in 2019 and operates from Paris, so GDPR applies directly to the customer photography a marketplace pushes through it — a more meaningful guarantee here than in prompt-only generation, because the inputs are real images of real products and sometimes of real people.
The consumer app is the better-known half, but the API is the half that belongs in this list: sellers, marketplaces and PIM systems calling it per image as products are onboarded. Pricing runs from a free tier through Pro plans from about €12.99 a month, with API usage billed per image.
What PhotoRoom does well
- Background removal and AI background generation
- Built for high-volume catalogue processing
- French company, GDPR applies directly
- Per-image API pricing
- Free tier for evaluation
Where PhotoRoom falls short
- Editing rather than general image generation
- No video generation
- No open weights or self-hosting
- Consumer app is the primary product
Standout feature. Product photography at catalogue scale from an EU company, rather than one-off creative generation.
Malaga, Spain
Founded 2010
Subscription + API credits
Best for: Content teams that need generation and licensed stock from one vendor
Freepik AI is the only provider here that combines generative models with a licensed stock library, and for content teams that combination removes a whole vendor relationship: search the library of over 250 million assets, or generate what is not in it, through the same API account.
Freepik Company S.L. operates from Malaga and has been in the stock content business since 2010, which is why the licensing side is more mature than at any AI-first competitor. Mystic, its image generation engine, is built in-house rather than being a wrapper around someone else's model, and the API also fronts a catalogue of video models.
The caveat is access: Freepik added API endpoints in late 2025, and availability is oriented towards enterprise plans with a narrower model selection than the dedicated developer platforms. If you want a self-serve key in five minutes, start elsewhere; if you want generation and cleared stock under one contract, this is the one that offers it.
What Freepik AI does well
- Generation plus a 250M+ asset licensed library
- Mystic is an in-house model, not a wrapper
- Video models available through the same API
- Spanish company, GDPR applies directly
- Mature licensing from fifteen years in stock content
Where Freepik AI falls short
- API access oriented towards enterprise plans
- Narrower model selection than dedicated platforms
- No open weights or self-hosting
- Stock library is the primary business, API the newer half
Standout feature. Generated and licensed stock imagery under one contract — no other provider in this category offers both.
Ukraine
Credits
Best for: Catalogues that need thousands of images cleaned up rather than invented
Claid.ai is the developer API from Let's Enhance, the Ukrainian company known for AI upscaling. Where Let's Enhance is the app you drag a photo into, Claid is the endpoint you point a catalogue at: upscaling, restoration, background generation for product photography and image-to-video, built for volume rather than one-off creative work.
Claid is one of the APIs in this category aimed squarely at improving images you already have rather than inventing new ones, alongside Deep-Image.AI and the real-estate-specific Autoenhance.ai. For an ecommerce catalogue of supplier photos in eleven different sizes and lighting conditions, that is a more useful problem to solve than text-to-image, and consistency across the batch matters more than any single spectacular result.
The jurisdiction needs stating plainly: Ukraine is neither an EU or EEA member nor covered by an EU adequacy decision. Ukrainian data protection law has been converging on the GDPR as part of the accession process, but the adequacy finding does not exist yet, so transfers there need their own safeguards — a real consideration if the images are of identifiable people rather than of products.
What Claid.ai does well
- Bulk upscaling, restoration and background generation
- Image-to-video from existing photographs
- Built for catalogue-scale jobs, not single requests
- Tuned for photographic fidelity over creative range
- Credit pricing with volume rates
Where Claid.ai falls short
- Ukraine has no EU adequacy decision
- Editing and enhancement rather than generation
- Smaller company than the rest of this list
- No open weights or self-hosting path
Standout feature. One of the few APIs here built to make thousands of existing photographs consistent rather than to invent new ones.
Harrow, London, United Kingdom
Founded 2023
Pay-per-use: image generation about $0.0006–$0.24/image; video from $0.09/sec (720p)
Best for: Teams that want the cheapest, broadest model catalogue on one endpoint
Runware is an inference platform rather than a model builder: one API in front of more than 400,000 image, video, audio and 3D models, from open weights like FLUX and Stable Diffusion to proprietary models from OpenAI, Google and ByteDance.
Where Nebius AI Studio in this list hosts a curated set of open models, Runware's pitch is breadth and price — switching models is a string change, and custom inference hardware ("Sonic Pods") is built to undercut market rates on a pay-per-request basis.
Runware Ltd was incorporated in June 2023 and is registered in Harrow, Greater London, so the UK adequacy decision covers transfers. The company is explicit in its own terms that "the Services are hosted in the United States, Germany and Romania" — a materially different residency promise from the EU-only or UK-only processing most other tools in this category describe, and worth weighing against the low prices.
Pricing is usage-based throughout: image generation typically runs $0.0006 to $0.24 per image depending on model and resolution, and video starts around $0.09 to $0.12 per second at 720p. For a product that needs many different models behind one integration and cares more about cost than about a guaranteed EU-only data path, that combination is hard to beat on price alone.
What Runware does well
- 400,000+ image, video, audio and 3D models behind one API
- Pay-per-request pricing, often well below other hosted APIs
- Open-weight models (FLUX, SDXL) alongside proprietary ones
- UK-registered company with a public VAT number
- Batch requests across modalities in a single call
Where Runware falls short
- Processing is spread across the US, Germany and Romania, not EU-only
- A model host, not a model builder, so no distinctive model of its own
- Founded in 2023 — the youngest company in this category
- Calls to proprietary models likely route onward to those vendors' own infrastructure
Standout feature. The broadest and cheapest model catalogue in this category, at the cost of a guaranteed EU-only data path.
Szczecin, Poland
Free tier, plus one-time credit packs or subscriptions, from free to enterprise
Free tier available
Best for: Catalogues that need upscaling, background work and generation chained in one call
Deep-Image.AI is a Polish API built around chaining image operations rather than picking one: upscale, denoise, deblur, correct colour, remove or generate a background and even generate new content, in a single request billed as one processed-image credit for compatible workflows. Its packshot generator and document upscaler (up to 300 megapixels) target the same high-volume, low-glamour work that PhotoRoom and Claid.ai serve, from a smaller and less established company.
The operating entity, Deep-Image.AI Sp. z o.o., states in its own terms that they are governed by the laws of Poland, and its privacy policy gives a registered address in Szczecin — so GDPR applies directly and there is no adequacy question to weigh, unlike the UK-based tools in this category.
Pricing runs on credits, bought as one-time packs or through a recurring subscription, with a free tier to evaluate the tools first; the public pricing page itself only promises a range "from free to enterprise" rather than a fixed price list, so exact rates are best checked live before committing volume.
What Deep-Image.AI does well
- Chains upscale, denoise, background removal and generation in one API call
- Packshot generator and document upscaler built for volume
- Polish company, GDPR applies directly
- Free tier and one-time credit packs, no subscription required
- Python client and public API documentation
Where Deep-Image.AI falls short
- Smaller and less established than PhotoRoom or Claid.ai
- Exact pricing sits behind a live dashboard rather than a published price list
- No open weights or self-hosting path
- No video generation
Standout feature. The only API here that chains upscaling, background removal and generation into a single processed-image credit.
Luxembourg
Founded 2023
Free tier; paid plans from about €8.25/month to €165.83/month (billed yearly), plus enterprise
Free tier available
Best for: Creative teams that want a generation platform with API access included
LetzAI is closer to Freepik AI than to the model builders in this list: a generative platform for images and video, with Canvas and Boards for team collaboration, that curates more than thirty foundation models from providers including Google, ByteDance and Black Forest Labs alongside thousands of community-trained custom models — and bundles API access into every paid plan rather than selling it separately.
The operating entity is Depixit S.à r.l., and the company describes itself directly in its own site copy as "a Luxembourg company" with personal data "handled under the GDPR." It was founded in 2023 by Misch Strotz and Karim Youssef, and lists GCore's Luxembourg infrastructure and the EuroHPC Meluxina supercomputer among its compute partners, alongside participation in the EU AI Factory programme.
Pricing runs on credits across four subscription tiers, from about €8.25 a month (Beginner, billed yearly) to €165.83 a month (Ultra), plus a free tier and top-up packs; Enterprise plans add a dedicated Model Training API. The credit cost per generation varies sharply by model, from budget image models to frontier ones, which adds a layer of comparison shopping a flat per-image rate does not.
What LetzAI does well
- Luxembourg-founded, GDPR applies directly, part of the EU AI Factory programme
- API access included on every paid plan
- Curates 30+ foundation models plus thousands of community-trained ones
- EuroHPC (Meluxina) and GCore Luxembourg among its infrastructure partners
- Transparent per-model credit pricing
Where LetzAI falls short
- Founded in 2023, one of the youngest companies in this category
- Curates other vendors' models rather than building its own
- Credit system adds a layer of pricing complexity versus a flat per-image rate
- Built as a creator platform with an API bolted on, not an API-first product
Standout feature. A Luxembourg-registered creative platform that bundles API access into every plan rather than selling API access alone.
London, United Kingdom
Founded 2020
Free watermarked preview, no card required; pay-per-image or monthly subscription; enterprise on request
Free preview, no card required
Best for: Real estate platforms and agents cleaning up listing photos at volume
Autoenhance.ai is the narrowest tool in this category by design: an API built specifically for real estate photography, covering sky replacement, HDR merge, perspective correction, window pulling and virtual-staging-style room changes, sold as both a consumer app and a developer API rather than a general image editor.
Autoenhance.ai Ltd has been registered in London since July 2020, one of the longer-established companies added in this update, so the UK adequacy decision covers transfers the same way it does for Stability AI, Recraft and Synthesia elsewhere in this list.
Pricing is pay-per-image or a monthly subscription, with a free watermarked preview that needs no account or card, and enterprise plans for high-volume portals; exact rates run through a live calculator on the site rather than a fixed table, so they were not carried over as specific numbers here.
What Autoenhance.ai does well
- Purpose-built for real estate: sky, HDR, perspective and window corrections
- Available as both a consumer app and a developer API
- Free watermarked preview, no account or card required
- UK company established since 2020
- Pay-per-image option avoids a subscription commitment
Where Autoenhance.ai falls short
- Vertical-specific — not a general image generation or editing API
- UK rather than intra-EEA processing
- No video generation
- Pricing sits behind a calculator rather than a published price list
Standout feature. The only API here built specifically for real estate photography rather than general product or creative imagery.
