D-ID
D-ID is a digital human platform that turns scripts, documents and photos into multilingual AI avatar videos, and deploys real-time visual agents that listen, answer and act. Available as a self-service studio or an API.
What is D-ID?
D-ID is a digital human platform built by De-Identification Ltd., an Israeli company founded in 2017 whose first work was an algorithm designed to protect faces against recognition systems — hence the name. The product line has since moved from privacy to generation, and now covers two distinct jobs.
The first is asynchronous video. The Creative Reality Studio, a self-service application available on desktop and mobile, combines deep-learning face animation with LLM text generation and text-to-image. You pick a face three ways: a ready-made presenter, an uploaded photograph, or a portrait generated by a Stable Diffusion-powered prompt. You give it a voice three ways: a typed script, an uploaded recording, or a cloned voice. The V4 Expressive avatars add facial expression, accurate lip-sync and delivery matched to the tone of the message, while Video Translate localises an existing recording while keeping the speaker's voice and visual identity.
The second is real-time interaction. Visual agents hold spoken or written conversations, keep persistent memory across sessions, ground their answers in approved documents and business systems, display images, forms and quizzes mid-conversation, and trigger external workflows. Agentic Videos let a viewer interrupt a video and ask questions. Delivery runs over WebRTC streaming, an API, client SDKs or ready-made embeds, and expressive avatars can be self-hosted inside a customer's own cloud.
Scale claims are substantial: more than 200 million avatar videos generated and over 280,000 developers, with PepsiCo, Fidelity, J.P. Morgan, SoftBank, NTT, Deutsche Telekom, PwC, Deloitte and AXA named as customers. D-ID acquired simpleshow in September 2025.
The constraints are equally concrete. Videos are capped at five minutes; images must be JPEG or PNG under 10 MB; audio under 10 MB and five minutes; output is MP4 at up to 1280x1280 for standard presenters and 1080p for premium ones outside the Lite plan. Every video carries a watermark that no plan removes — Enterprise can only customise it — which the company frames as a transparency commitment about synthetic content.
What it does
- Generate a talking avatar video from a script, brief, deck or document
- Deploy a real-time visual agent that listens, answers and triggers workflows
- Translate an existing video while preserving the speaker's voice and appearance
- Create a reusable personal avatar from a photo or a short video recording
- Clone a voice and deliver the same message across 119 languages
- Turn a video into an Agentic Video that answers viewers' questions during playback
- Embed avatars and agents in a product through API, SDK, embeds or WebRTC
When to use D-ID / When not to
A quick filter to help you decide if D-ID is the right fit.
When to use D-ID
- Learning and development teams producing multilingual training videos at scale
- Marketing and communications teams personalising video content across languages and channels
- Customer experience teams deploying real-time conversational agents with a human face
- Developers embedding avatars or visual agents through an API, SDK or WebRTC stream
- Regulated organisations that need ISO 27001, 27018, 42001 and SOC 2 credentials from a video vendor
When not to use D-ID
- Anyone who needs watermark-free output, since no plan removes the watermark
- Trial or Lite subscribers intending commercial use, which the terms forbid
- Teams needing videos longer than five minutes per render
- Buyers who must compare prices before signing up, as no amount is published in readable form
- Projects animating animals, cartoons or anime characters, which face detection rejects
How to use D-ID
A typical end-to-end flow, from setup to results.
- Create an account on the D-ID studio and start the free trial
- Choose a face: a stock presenter, an uploaded portrait, or a text-to-image prompt
- Respect the image guidelines — facing camera, medium shot, neutral expression, closed mouth, head at least 200x200 pixels, even lighting, no hats, sunglasses or masks
- Give the avatar a voice by typing a script, generating one, uploading a recording, or cloning a voice with recorded consent
- Compose the scene with layers for backgrounds, video and text, and pick a wide, square or vertical canvas
- Set the expression and add half-second pauses where the delivery needs them
- Generate the video, remembering that duration is rounded up to the next 15-second step
- For the API, generate a key from the studio Account page with credits in the balance, then follow the Developer Hub
- For an agent, define its role, tone and boundaries, connect approved knowledge sources, then deploy it as an embed or through the SDK
- Note that layers, expressions and larger credit packs are unavailable in the mobile app
Pros & Cons
Pros
- Unusually broad language coverage, with 119 languages in the studio
- One platform for both asynchronous video and live conversational agents, rarely combined
- Deep compliance base: ISO 27001:2022, 27017, 27018, 42001, 27799 and SOC 2
- You keep ownership of your submissions, output and animations
- Applicative data is not accessed for model training
- Self-hosting available for sovereignty or latency constraints
- Mature vendor: founded in 2017, named enterprise customers, 24/7 support for studio and API
Cons
- No price is readable without JavaScript, as both pricing pages are client-side applications
- The watermark cannot be removed on any plan, only customised on Enterprise
- Trial and Lite are limited to non-commercial use by the terms
- Videos are capped at five minutes and credits expire at the end of each month
- No published DPA, no Article 27 EU representative and no named Data Protection Officer
- Applicative data is hosted in the United States with no documented European region option
- Voice cloning is rationed by plan and several studio features are missing on mobile
Pricing & Plans
A free trial is available: the Trial plan runs for two weeks from registration, with limited features, non-commercial use and a full-screen watermark. Whether a permanent free plan exists could not be established. No entry price can be quoted: both the Studio and API pricing pages are rendered client-side by JavaScript, and no amount or currency appears in the served HTML, in the published FAQ, or anywhere else on the site. Billing is monthly or annual and credit-based, one credit covering up to fifteen seconds of video, with streaming credits charged at half price through the API. Credits are renewed monthly and unused ones are lost. Prospective buyers must open the pricing page in a browser or contact sales.
- two weeks from registration
- limited features
- non-commercial use
- full-screen watermark
- D-ID logo watermark
- Premium presenters not supported
- non-commercial use only
- generic AI watermark
- one cloned voice
- Premium presenters at 1080p
- generic AI watermark
- three cloned voices
- manual review on request
- own moderation allowed
- customisable but non-removable watermark
- customisable number of cloned voices
- one cloned voice
- three cloned voices
Data, GDPR & hosting
A consolidated view of how D-ID handles your data.
GDPR overview
Implementation is concrete but partial. The privacy policy sets out a GDPR legal basis for each category of data — legitimate interest, performance of a contract, or consent — and lists data subject rights, exercised through support@d-id.com. Transfers outside the EU and EEA rely on European Commission adequacy decisions under Article 45, explicitly citing access from Israel, or on Standard Contractual Clauses. The trust centre presents ISO/IEC 27018:2019 as covering compliance with data protection laws including the GDPR, alongside ISO 27001:2022, 27017:2015, 42001:2023, 27799 and SOC 2. Californian provisions are also included. Three gaps are worth noting: no Article 27 EU representative is designated, no Data Protection Officer is named, and no Data Processing Agreement is published or offered on request in the pages reviewed.
Who owns the data?
Under D-ID's Products Terms of Use, you keep ownership of everything you bring and everything you generate: user submissions, output and animations remain yours. D-ID and its licensors keep ownership of the software, the underlying models and the intellectual property in them, including the D-ID marks and the synthetic-content watermark. Commercial use, defined as creating and reselling animations or derivative works to third parties, is permitted under the End User Agreement, but the Trial and Lite plans are restricted to non-commercial use. Voice cloning requires a recorded voice consent inside the submitted audio, so rights over a cloned voice rest on that consent rather than on the platform.
Reuse rights
Yes, within limits. Generated animations can be reused without asking D-ID for permission: the End User Agreement grants a right of commercial use, defined as creating and reselling animations or derivative works to third parties, subject to the agreement's terms. Two restrictions matter. The Trial and Lite plans allow non-commercial use only, so reuse under those plans is not licensed. And every video carries a watermark that cannot be removed, only customised on Enterprise, so any redistribution keeps signalling that the content is synthetic. On the vendor's side, applicative data is not accessed for model training while stored, and personal data serves to run and secure the service, meet legal obligations, and support marketing where consent was given.
Data retention & training
Hosting summary
Applicative data and the software run on AWS with servers located in Oregon, United States. The website side relies on Salesforce, HubSpot and ZenDesk, also on US servers. The privacy policy names these providers explicitly, which is more disclosure than many vendors offer, though there is no dedicated subprocessor page. Data at rest sits in S3 storage with transparent data encryption; traffic is encrypted with TLS 1.3; servers and storage sit behind a firewall and a web application firewall, and workstations are encrypted and monitored with endpoint detection and response. Transfers outside the EU and EEA rely on European Commission adequacy decisions under Article 45 — access from Israel is the example given — or on Standard Contractual Clauses. No European hosting region is documented and no Data Processing Agreement is published. Customers with sovereignty or latency requirements can self-host the expressive real-time avatars inside their own cloud, with scaling driven by metrics they define, which is the only documented route to keeping the workload off D-ID's US infrastructure.
Things to keep in mind
Risks and trade-offs to weigh before adopting D-ID.
- Synthetic likeness is easy to misuse: an avatar built from someone's photograph or a cloned voice needs their genuine consent, not just a ticked box
- The watermark is the only visible signal that a video is synthetic, and it stays on every plan — treat any wish to hide it as a warning sign about the project itself
- Trial and Lite are non-commercial by contract, so using them for client work is a licensing breach rather than a grey area
- Applicative data sits in the United States with no documented European region, and there is no published DPA to sign
- Deleting an account is irreversible: videos, agents and stored data go with it, and unused credits expire every month regardless
- Automated moderation can reject an image, script or audio file, with manual review reserved to the higher plans, which can stall a deadline
- Leaning on avatars for every message can flatten a brand's voice and let teams skip the harder work of having something worth saying
Setup & Integrations
Technical difficulty
Low for the studio: create an account, pick a presenter, type a script and generate. The API page claims zero technical knowledge is required, and the office integrations install as a PowerPoint add-in or a Canva app. Moderate for the API: generate a key from the Account page with credits in the balance, then follow the Developer Hub. Higher for real-time work — WebRTC streaming, client SDKs, connecting an agent to your own knowledge sources and models — and highest for self-hosting avatars in your own cloud. Support is offered around the clock to studio and API customers.
Deployment
Integrations
Supported languages
Behind D-ID
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What can I actually make with D-ID?
How many languages does it support?
Can I remove the watermark?
Who owns the videos I generate?
Is my data used to train D-ID's models?
How long is my data kept?
What are the technical limits?
How much does it cost?
Is there an API?
What is the minimum age?
Should you pick D-ID?
D-ID has been in this field since 2017, longer than most of the avatar tools it competes with, and it shows in two places: the breadth of the platform and the depth of its compliance paperwork. Few vendors let you produce a scripted multilingual video in the morning and embed a live conversational agent in a product the same afternoon, and fewer still bring ISO 27001, 27017, 27018, 42001, 27799 and SOC 2 to the conversation. For a buyer whose security review is the real gatekeeper, that combination is the argument.
Two things temper it. The first is commercial opacity: both pricing pages are JavaScript applications, so no amount is readable in the served page, and a prospect cannot compare D-ID with an alternative without opening a browser or talking to sales. The second is the watermark, which no plan removes and only Enterprise can customise. D-ID frames it as an ethical commitment to signalling synthetic content, which is defensible, but it remains a hard constraint on any polished external deliverable.
The legal posture is better than average on ownership — you keep your submissions, output and animations, and applicative data is not used to train models — and weaker than average on GDPR formalities: no published DPA, no Article 27 representative, no named DPO, and applicative data hosted in Oregon with no documented European region.
The fit is clear enough. If you industrialise multilingual video for training, marketing or support, or you want a human face on an agent inside your own product, D-ID is a serious candidate, especially where certifications decide the purchase. If you need watermark-free output, videos beyond five minutes, commercial use on an entry plan, or a price you can check before signing up, look elsewhere first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.