D-ID is a generative video and digital-human platform. Its Creative Reality Studio turns a script or audio track plus a stock, photo or custom video avatar into a presenter video. Developer APIs support prerecorded “talks,” video translation, real-time streaming avatars and interactive visual agents. The platform is used for training, sales, support, localization, education and embedded conversational experiences.
The production shortcut is real: a team can update a spoken explainer without booking a studio and can localize it into multiple languages. The governance burden is also real. A face and voice are identity signals. The person shown may not have agreed to a particular script, language, advertisement or automated interaction. Upload capability is not consent, and a paid subscription is not a license to impersonate someone.
Choose the correct D-ID product surface
| Surface | Input | Output | Best fit |
|---|---|---|---|
| Creative Reality Studio | Script/audio, avatar, voice and visual settings | Rendered presenter video | Non-developers producing training, marketing or explainers |
| Talks / video API | Image/video avatar plus text or audio | Asynchronous rendered clip | Templated video at application scale |
| Video Translate | Existing speaker video and target language | Translated speech with lip synchronization | Localized campaigns, training and product media |
| Visual Agents | Avatar, model, voice, instructions and knowledge | Interactive streaming conversation | Guided support, sales or learning interfaces |
| Agentic Videos | Video narrative plus interactive agent configuration | Video that can answer viewer questions | Interactive explainers where linear playback is insufficient |
Do not use a real-time agent when a prerecorded, reviewed video is enough. Real-time generation adds open-ended language, session costs, retrieval risk, latency and moderation. Likewise, translation is not merely a rendering feature: it creates a new performance in a language the original speaker may not understand.
End-to-end trust workflow
rights-cleared face + voice + script + purpose
|
consent record
scope / language / channel / duration / revocation
|
Studio / Translate / API / Agent
|
moderation + generation
|
.------------------+------------------.
v v v
identity QA language QA technical QA
likeness/voice meaning/pronounce lip-sync/export
'------------------+------------------'
v
disclosure + watermark + human approval
|
publish / monitor / revoke / delete
A release checklist should link the generated asset to the consent record, source hashes, script version, target languages, D-ID plan/API version, reviewer, disclosure and expiry. That allows a team to unpublish all variants when a speaker withdraws consent or a claim becomes outdated.
Studio versus API plans
D-ID sells Studio and API subscriptions separately. Studio tiers have historically included Trial, Lite, Pro, Advanced and Enterprise; API pricing currently shows Trial, Build, Launch, Scale and Enterprise. Exact prices and allowances change, so verify the live page at purchase. More important are license, watermark, avatar, voice-clone, storage and processing boundaries.
| API plan (review date) | Published offline video | Published streaming | License / branding highlights |
|---|---|---|---|
| Trial | Up to 3 minutes | Up to 10 minutes | Personal; full-screen watermark; standard voices |
| Build | Up to 16 minutes | Up to 32 minutes | Personal; D-ID watermark; own S3 option |
| Launch | Up to 45 minutes | Up to 90 minutes | Commercial; AI watermark; one voice clone; premium voices |
| Scale | Up to 200 minutes | Up to 400 minutes | Commercial; custom logo; three voice clones |
| Enterprise | Custom | Custom | Custom branding, collaboration, security and service options |
The live API page in July 2026 displayed annual-equivalent pricing of $14.40/month for Build, $35/month for Launch and $138.60/month for Scale, with stated discounts; monthly billing differs. These figures are time-sensitive and not a quote. Confirm taxes, reasonable-use limits, credit bundles and the actual checkout screen.
How credits and minutes work
One standard credit covers up to 15 seconds of generated or translated video. Duration rounds up by 15-second intervals: a 40-second clip uses three credits; a 70-second clip consumes 75 seconds. Studio and API usage can draw from the same account balance. Unused monthly credits do not roll over.
| Workload | Meter | Budget trap | Control |
|---|---|---|---|
| Rendered video | 15-second credit increments | Many short clips waste rounded intervals | Batch script segments deliberately |
| Video translation | Also duration/credits | Every target language multiplies usage and review | Pilot one language and one representative minute |
| Streaming/Agent speech | Published rates use smaller credit increments; plan minutes differ | Verbose responses consume budget and frustrate users | Limit answer length and session turns |
| Retries | May generate chargeable variants | Pronunciation fixes and visual artifacts multiply cost | Approve script/phonetics before full render |
| Unused allowance | Expires each billing period | Annual contract still issues credits monthly | Forecast seasonal volume from historical scripts |
For Visual Agents, help materials state that a generated-video reply of up to 15 seconds consumes 0.5 credits, with additional 15-second intervals charged similarly. Confirm the current plan because interactive and offline allowances are presented differently. The agent stops responding when credits run out, so production applications need balance monitoring and a text/static fallback.
Commercial rights are plan-specific
D-ID’s help center says Trial, Lite and Build outputs are for personal use, while Pro, Advanced, Enterprise, Launch and Scale provide commercial use, assuming the user owns rights to uploaded image, audio and text. That assumption is crucial. D-ID grants rights under its product terms; it cannot grant publicity, copyright, trademark, employment or voice rights you never obtained from the subject or asset owner.
| Asset/right | Evidence to retain | Question before reuse |
|---|---|---|
| Person’s likeness | Signed release identifying AI animation and distribution channels | Does it cover this script, language, territory and paid advertising? |
| Voice recording/clone | Voice-specific consent and source provenance | Can the voice say newly generated text or only approved scripts? |
| Script | Author/license, source citations and legal approval | Are claims current and permitted in the target market? |
| Stock avatar/voice | D-ID plan/license terms at generation date | Are commercial use and sensitive categories allowed? |
| Music/images/logos | License and brand authorization | Does localization or paid media exceed the license? |
Watermarks and disclosure
Official guidance describes a full-screen watermark on Trial, a D-ID mark on Lite/Build, a generic AI mark on Pro/Launch, custom-logo capability on Advanced/Scale, and removal or replacement on Enterprise. Current plan naming can differ between Studio and API. A custom logo or watermark removal does not eliminate disclosure duties.
Disclose material synthetic media where viewers could reasonably believe a real person spoke the words. Place disclosure in the video and surrounding page, not only metadata. For advertisements, political or financial messages, education assessments and customer service, review local law and platform rules. Preserve a machine-readable generation record where possible.
| Context | Minimum disclosure pattern | Additional safeguard |
|---|---|---|
| Internal training | “AI-generated presenter” on opening/end card | Named content owner and update date |
| Public marketing | Persistent or clearly visible synthetic-media label | Consent/contact path and substantiated claims |
| Translated real speaker | State that voice/lip movement were AI-translated | Speaker approves final target-language meaning |
| Interactive agent | Introduce itself as AI before conversation | Human escalation and no deceptive emotional claims |
| High-impact advice | Avoid avatar as authority substitute | Qualified human review and explicit limitations |
Voice cloning and personal avatars
Voice cloning can make arbitrary text sound like a real person. Obtain explicit, revocable consent that names acceptable subjects, languages, channels and duration. Do not accept a checkbox from an uploader as the only evidence when a business is cloning employees, customers or public figures. Add liveness or identity verification where risk warrants it.
- Store the consent artifact separately from the production API key.
- Restrict which users can create a clone and which applications can call it.
- Block financial instructions, credentials, emergency messages and unauthorized endorsements.
- Notify the subject about new campaigns and translated languages.
- Provide a fast revocation workflow that disables the clone and locates published outputs.
- Red-team phonetic prompts, cross-language impersonation and attempts to bypass moderation.
Translation QA
| Layer | Check | Typical failure |
|---|---|---|
| Meaning | Back-translate and compare claims, conditions and negation | Fluent sentence changes obligation or product promise |
| Terminology | Apply approved glossary for product, legal and technical terms | Brand or regulated term translated inconsistently |
| Pronunciation | Native reviewer checks names, acronyms and numbers | Correct text but unusable spoken delivery |
| Voice identity | Subject confirms the synthetic voice is acceptable in that language | Accent or tone implies a false identity |
| Lip sync | Inspect close-ups, occlusion, facial hair and fast speech | Mouth artifacts distract or mislead |
| On-screen text | Localize captions, diagrams and calls to action separately | Spoken language changes while visuals remain original |
API implementation and reliability
D-ID’s API key is presented once and uses an API username/password form through the Authorization header. Store it in a secrets manager, never a browser/mobile client. A rendered-video request is asynchronous: applications submit a job, poll or receive completion information, then handle an output URL. Build idempotency at your layer so retries do not create duplicate chargeable renders.
validate rights + script -> reserve internal job id
|
submit once to D-ID
|
pending -> processing -> done / error / moderation block
|
copy approved output to controlled storage
|
QA + disclosure -> publish
|
expiry / revocation -> unpublish + delete
| Failure | Application behavior |
|---|---|
| 401/rotated key | Stop, alert and rotate; never retry with logged credentials |
| 429/quota | Back off with jitter; check balance and provide queue estimate |
| Moderation rejection | Do not automate evasion; route to content owner/manual review |
| Render timeout | Poll existing job before resubmitting |
| Partial translation/localization | Fail the release as a set; do not publish mismatched variants |
| Credit exhaustion | Fall back to prerecorded/text interaction and notify operator |
Visual and audio acceptance test
| Dimension | Pass criterion |
|---|---|
| Identity | Authorized subject/avatar, no unintended resemblance or face distortion |
| Speech | Words, numbers, names, pauses and emphasis match approved script |
| Sync | No material mouth/voice mismatch at normal speed and target devices |
| Framing | Eyes, mouth, logo, captions and disclosure remain inside safe areas |
| Accessibility | Accurate captions, transcript, sufficient contrast and non-audio alternative |
| Export | Resolution, codec, audio level and playback work on the actual channel |
| Truthfulness | Claims and sources are current; synthetic presenter does not imply personal endorsement |
Privacy, storage and deletion
Uploaded face images, videos and voice recordings can be biometric or sensitive personal data depending on jurisdiction. D-ID’s privacy materials distinguish Studio content, which can persist until the user erases it, from API application data with job-specific handling. Current pricing says users may connect their own S3 storage on some API plans. Own storage does not remove upstream processing or logs.
Map every copy: source device, D-ID account, API job, output URL, your object storage, CDN, editor, campaign platform, backups and analytics. Account deletion removes access and stored items under the documented process, but a production deletion test should confirm API/Studio assets, agent knowledge, avatars, voice clones and downstream copies. Free inactive-account deletion after six months is not a retention policy for an organization.
When D-ID is a good fit
| Use case | Fit | Reason |
|---|---|---|
| Frequently updated training | Strong | Script changes without a reshoot; versioning matters |
| Localized product explainers | Strong with native review | Translation and lip sync reduce production work |
| Interactive FAQ concierge | Conditional | Visual presence may help, but text/chat can be faster and cheaper |
| Executive impersonation without explicit consent | Do not use | Severe identity, fraud and reputation risk |
| Emotionally sensitive counseling | Weak/high risk | Human-like avatar can create deceptive trust |
| Cinematic generative video | Weak | D-ID specializes in presenters; use broader video models/tools |
Alternatives
| Alternative | Best when | Compare |
|---|---|---|
| HeyGen | Marketing-friendly avatars and translation workflow | Avatar quality, consent controls, languages, API and pricing |
| Synthesia | Enterprise learning, templates and governed stock avatars | Collaboration, security, custom avatar policy and localization |
| DeepBrain AI | Studio-style AI presenter and broadcast workflows | Realism, language coverage and interactive API |
| Tavus | Personalized video and conversational replicas | Replica consent, latency and developer integration |
| Traditional presenter/voice actor | Trust, emotional nuance and endorsement are central | Higher cost/time but authentic performance and clearer rights |
| Audio + slides/screencast | The avatar adds little informational value | Lower identity risk, cost and uncanny-valley distraction |
FAQ
Can D-ID animate any photo?
No. Images can fail face detection or moderation, and rights/consent are required even when the system accepts the upload.
Can I use Trial or Build output commercially?
Official guidance says Trial, Lite and Build are personal-use tiers. Commercial rights are associated with Pro, Advanced, Enterprise, Launch and Scale, assuming you own all input rights.
How are video credits rounded?
Standard generation rounds duration up in 15-second intervals. A 40-second clip uses three credits. Verify interactive-agent rates separately.
Can the watermark be removed?
It depends on plan. Some tiers show D-ID or generic AI branding; higher tiers support a custom logo or removal. Ethical/legal disclosure may still be required.
Does D-ID provide consent for stock or uploaded people?
D-ID provides its platform and terms; the uploader remains responsible for rights to uploaded image, voice and script and for the intended use.
What should be tested before API launch?
Idempotency, authentication, rate/credit exhaustion, moderation failures, output expiry/storage, consent lookup, disclosure, accessibility and deletion.
Sources and verification
- Official D-ID product site
- Official Studio pricing and credit FAQ
- Official API plans and current allowances
- Official developer documentation
- Official API authentication guidance
- Official commercial-use rights by plan
- Official watermark behavior
- D-ID ethical manifesto
- D-ID privacy policy
Last reviewed July 26, 2026. Plans, prices, credits, moderation and legal requirements change. Confirm live terms and obtain qualified legal advice for identity-sensitive or regulated uses.
