ModelsLab
ModelsLab is a unified generative AI API for developers: one key covers image, video, voice, 3D and LLM models, plus dedicated GPU deployments. There is no free tier — paid plans start at USD 21 per month.
What is ModelsLab?
ModelsLab sells a single door onto generative AI. One API key, one request shape, and a catalogue that spans image, video, audio, 3D and large language models — moving from one model to another is a matter of changing the model_id field rather than rewriting an integration. The infrastructure stays on the vendor's side: no GPU to rent, no deployment to size, no model-by-model configuration.
The catalogue mixes open-weight families such as Flux, SDXL, Stable Diffusion and Qwen with closed third-party models from Bytedance, Alibaba, Black Forest Labs, MiniMax, LTX, Tencent, ElevenLabs and Google. How large it is depends on which page you read: the homepage counter shows 201 models, its meta description says 1,000+, the pricing page and the structured data say 10,000+, and llms.txt claims 100,000+. Audience figures diverge just as widely, from 3,600 developers on the pricing page to 300,000 in llms.txt. The platform also advertises 500 million API requests served and 99.9% uptime.
Money works on two floors. A monthly subscription buys API calls and concurrency — USD 21 for 3,250 calls, USD 47 for 10,000, and USD 149 for unlimited generation on the open-source models ModelsLab hosts itself. Premium third-party models are billed per call from a prepaid wallet on every plan, the unlimited one included. Above that sits a separate dedicated-GPU offer, from USD 249 to USD 1,999 a month, where capacity is isolated, rate limits disappear and outputs can land in the customer's own S3 bucket.
What sets the product apart is how deliberately it courts machines. There are MCP servers, a command-line client, installable agent skills, a published OpenAPI contract, markdown versions of every page, and a control plane that lets an autonomous agent create an account, pay through a Stripe link, issue its own key and start generating without a browser ever opening. Humans are served through plugins for Photoshop, Premiere Pro, After Effects, DaVinci Resolve, Figma, Blender, Unity and Maya, and a Chrome extension.
The corporate picture is thinner. The terms name M Studio AI Inc, the homepage structured data says ModelsLab Inc., and the refund policy still refers to Stable Diffusion API.
What it does
- Generate images from a text prompt across Flux, SDXL, Qwen, Stable Diffusion and Nano Banana
- Produce text-to-video and image-to-video clips with synchronised audio via Seedance, Kling, LTX, MiniMax and Vidu
- Synthesise, clone and convert voices, dub videos, and generate music or sound effects
- Call large language models such as GPT-5, Gemini 2.5 Flash, DeepSeek and Qwen through the same endpoint
- Edit images with inpainting, outpainting, upscaling, extension, retouching and virtual try-on
- Train custom models and LoRAs, then serve them from a dedicated GPU alongside your own checkpoints
- Generate 3D assets from text or an image, plus avatars, headshots, interiors and floor plans
When to use ModelsLab / When not to
A quick filter to help you decide if ModelsLab is the right fit.
When to use ModelsLab
- Backend and full-stack developers who need image, video, audio and LLM generation behind a single API key
- Product teams shipping generative features without provisioning or maintaining GPUs
- Studios and creative professionals working in Photoshop, Premiere Pro, After Effects, Blender, Unity or Figma, where ModelsLab ships native plugins
- Builders of autonomous AI agents: the control plane lets an agent sign up, pay, issue a key and generate without a browser
- High-volume workloads chasing a predictable bill, through the USD 149 unlimited open-source plan or a dedicated GPU from USD 249 a month
When not to use ModelsLab
- Anyone who wants to try before paying: ModelsLab publishes no free tier, no free credits and no trial
- Non-technical users looking for a finished creative application rather than an API
- Teams that need a signed DPA, a published subprocessor list or a documented hosting region before they can sign
- Buyers who require a verifiable place of establishment: no postal address appears anywhere on the site
- European organisations uncomfortable with terms governed by Indian law and the exclusive jurisdiction of the New Delhi courts
How to use ModelsLab
A typical end-to-end flow, from setup to results.
- Create an account: generation requires an active subscription, so there is nothing to try beforehand
- Pick a plan — Basic at USD 21, Standard at USD 47, or Open Source Unlimited at USD 149 a month
- Copy the API key issued in your dashboard
- Browse the model catalogue and open the model you want; each page shows the exact request body to send
- Call the endpoint (/api/v6, /api/v7 or /api/v8) with the key passed in the request payload
- Swap models later by changing model_id alone — the rest of the integration is untouched
- Top up the prepaid wallet if you intend to call premium third-party models
- Install an official SDK if you prefer: pip, npm, composer, go get or dart pub
- For agents and terminals, install the CLI, connect an MCP server, or add the agent skills with npx
- Move to a dedicated GPU when volume justifies it: the request shape is unchanged, only the base URL differs
Pros & Cons
Pros
- One integration replaces several providers across five modalities
- Changing model is a parameter change, not a refactor
- Predictable cost at volume: unlimited open-source generation at USD 149 a month, or a flat dedicated GPU from USD 249
- Commercial ownership of generated output is stated explicitly in the terms
- 100% refund policy on monthly and yearly subscriptions, minus usage already consumed
- Unusually thorough machine-facing tooling — MCP, CLI, agent skills, OpenAPI and markdown pages
- Native plugins bring the same models into the main creative applications without writing code
Cons
- No free plan, no free credits and no trial: the platform cannot be evaluated without paying first
- Two-tier billing to reconcile — a subscription plus a separate prepaid wallet for premium third-party models
- Plan allowance resets every period and does not roll over, while the wallet balance is non-refundable
- Corporate identity is inconsistent: M Studio AI Inc in the terms, ModelsLab Inc. in the structured data, Stable Diffusion API in the refund policy
- No postal address anywhere, and the privacy policy still carries an unfilled placeholder where the registered office should be
- No DPA, no subprocessor list, no stated hosting country and no security certification
- The refund policy contradicts itself, and the terms have not been revised since June 2023
Pricing & Plans
There is no free plan and no free trial. ModelsLab states plainly that the platform is paid-only, with no free tier and no free credits. The lowest entry point is the Basic plan at USD 21 per month, which includes 3,250 API calls and five concurrent requests. Annual billing grants two months free. A prepaid wallet is required on every plan, including the unlimited one, to call premium third-party models, and dedicated GPU deployments start at USD 249 per month.
- 3
- 250 API calls
- 5 concurrent requests
- shared GPU
- support via Discord and email
- 10
- 000 API calls
- 10 concurrent requests
- priority developer support
- production workloads
- unlimited generation on self-hosted open-source models
- 15 concurrent requests
- team members
- priority GPU clusters
- 24/7 support
- dedicated 24 GB VRAM GPU (RTX 3090)
- up to 100 loaded models
- no rate limiter
- dedicated 48 GB VRAM GPU (RTX 6000 Ada)
- up to 500 loaded models
- dedicated 80 GB VRAM GPU (A100)
- up to 1
- 000 loaded models
- Prepaid wallet — billed per call
- outside any plan
- for premium third-party models
- two months free
Data, GDPR & hosting
A consolidated view of how ModelsLab handles your data.
GDPR overview
GDPR compliance is claimed rather than demonstrated. The privacy policy names the company as data controller, lists the six lawful bases and the full set of data-subject rights — access, rectification, objection, erasure, portability and withdrawal of consent — and points EEA users to their local supervisory authority. The terms add that EU personal data is processed only to perform the service and is never sold. A GDPR badge appears on the pricing page and the enterprise offer is described as GDPR-aligned. What is absent matters just as much: no Article 27 EU representative, no named data protection officer, no data processing agreement, no subprocessor list and no stated hosting location. Security measures are described as complying with Indian law, the GDPR and India's Information Technology Act 2000, and the policy itself is governed by Indian law.
Who owns the data?
The terms are unusually clear on ownership: users keep every right, title and interest, including intellectual property, in the images they generate, and ModelsLab states that generations are fully copyrighted to the user and free to use commercially. The enterprise page repeats the point for dedicated deployments, including output produced from customer-uploaded checkpoints, and adds that outputs can be delivered straight into the customer's own S3 bucket. Two reservations apply: individual closed third-party models can carry their own upstream licence terms, and the vendor may reuse any feedback or suggestion without obligation. For personal data, ModelsLab describes itself as the data controller.
Reuse rights
End users may reuse what they generate without asking permission. The terms state that generations are fully copyrighted to the user and free for commercial use, and the pricing FAQ confirms it for every paid plan. The limit sits upstream: closed third-party models routed through the platform may impose their own licence conditions, so ModelsLab advises checking the model page before building a product on one specific provider. The licence on the platform itself is narrower — you may integrate the services into your own applications, but you may not resell, sublicense, reverse-engineer or create derivative works of the API. On the personal-data side, the company says it does not sell, rent or trade personally identifying information, and the terms commit to processing EU personal data solely to perform the service.
Data retention & training
Hosting summary
ModelsLab does not say where data is hosted. No country, no region and no data-centre location appears on any page, and the site publishes no trust or security page. The only substantive statement comes from the enterprise offer: on a dedicated GPU deployment, customers can connect their own S3 bucket, in which case outputs never persist in ModelsLab's storage and are delivered through private signed URLs. Without that bucket, outputs sit in the vendor's own storage, which the comparison table names but does not locate. The privacy policy describes physical, electronic and procedural safeguards said to comply with Indian law, the GDPR and India's Information Technology Act 2000 — a jurisdictional signal rather than a hosting statement. Both the terms and the privacy policy are governed by Indian law, with exclusive jurisdiction in New Delhi. The website itself is served through Cloudflare, which says nothing about where processing happens. Buyers with data residency requirements will need to raise the question directly, because no document on the site answers it.
Things to keep in mind
Risks and trade-offs to weigh before adopting ModelsLab.
- You cannot evaluate the platform before paying, which pushes buyers into a subscription decision on marketing copy alone
- Model counts and user counts differ by two orders of magnitude between pages, so treat every headline figure as a claim rather than a measurement
- The prepaid wallet is non-refundable while subscriptions are refundable: money moved into the wallet is committed
- Cancelling deletes your data from the servers permanently, according to the refund policy — export anything you need first
- Generated media carries no stated provenance marking, so responsibility for deepfaked voices, synthetic likenesses and celebrity voice cloning rests entirely on the user
- Closed third-party models keep their own upstream licences, so commercial ownership is not uniform across the catalogue
- Contract and privacy documents are governed by Indian law with exclusive New Delhi jurisdiction, which makes any dispute costly to pursue from Europe
Setup & Integrations
Technical difficulty
Moderate for developers, low for creatives, unsuitable for non-technical buyers. The core product is an HTTP API: authenticate with a key, post a JSON body, read the result. Each model page publishes the exact request body, and official SDKs cover Python, TypeScript, PHP, Go and Dart, with samples in nine languages. A CLI installs in one command, MCP servers and agent skills plug into coding agents, and moving to a dedicated GPU is a base-URL change. Creatives can skip code entirely through the plugins for Photoshop, Premiere Pro, After Effects, DaVinci Resolve, Figma, Blender, Unity and Maya.
Deployment
Integrations
Supported languages
Behind ModelsLab
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement ModelsLab.
Frequently asked questions
Is there a free plan or a free trial?
What does the USD 149 unlimited plan actually cover?
Do I still need to fund a wallet?
Does my plan allowance roll over?
Can I use what I generate commercially?
Which modalities does the API cover?
How do I integrate it?
Can an autonomous agent sign up and pay by itself?
Where is my data hosted, and is a DPA available?
Which law governs the contract?
Should you pick ModelsLab?
ModelsLab makes a coherent promise: consolidate a scattered generative AI stack behind one key. The breadth is real — image, video, voice, 3D and language models sit behind the same request shape, and moving between them costs a parameter rather than a rewrite. The machine-facing tooling is genuinely ahead of the field, with MCP servers, a CLI, installable agent skills and a control plane an autonomous agent can drive from signup to first generation. For a team already past the exploration phase and running enough volume to feel per-request billing, the unlimited open-source plan and the dedicated GPU tiers offer something increasingly rare: a line item that does not move with traffic.
The reservations are about the company more than the product. Nothing can be tested before paying, which is an unusual ask in this market. Billing has two floors — subscription and prepaid wallet — and the split between self-hosted open-source models and metered third-party models needs to be understood before the first invoice, not after. More seriously, the legal surface is thin: three different company names across three documents, no postal address anywhere, a privacy policy still carrying unfilled placeholders, terms untouched since June 2023, a self-contradictory refund policy, and no DPA, subprocessor list, hosting location or security certification. Indian law and exclusive New Delhi jurisdiction complete the picture.
Who it suits: technical teams that have validated their need, want to collapse several vendors into one integration, and can absorb the procurement gaps. Who should look elsewhere: anyone whose purchasing process requires a signed DPA, a documented hosting region or a verifiable legal entity — and anyone who simply wants to try the thing first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.