Text scripts are synthesized by a connected speech provider before visual rendering. Video footage enrolls V3/V4 avatars; generation selects a saved avatar.
Spoken text excluding SSML tags. Total text/SSML field is capped at 40,000 characters; connected TTS providers may impose lower limits. Characters are not tokens.
The route selects an avatar or presenter ID, not a public immutable weight/checkpoint selector.
Access
POST https://api.d-id.com/expressives; use the returned job ID to poll status or receive a webhook, then retrieve the result URL. API credentials and eligible account access required.
Regions
No single public-cloud processing region or regional inference guarantee established for this route.
Data retention
Provider privacy policy describes commercially reasonable deletion of non-persisted API applicative data within 24 hours, Agents within 72 hours, and Insights within 14 days. Persisted assets, knowledge documents and Studio uploads have different lifetimes. Support/operational access is allowed; this is not default zero retention.Source for Data retention (opens in a new tab)
persistentData
persist=true assets, knowledge documents and Studio uploads persist until erased. While awaiting processing/deletion, applicative data is not accessed for model training; authorized support/operational access remains possible.Source for persistentData (opens in a new tab)
ownStorage
Provider policy permits signed S3 input/output URLs; when the result is directed to customer S3, the result is not stored in D-ID’s system. This is not a blanket promise of no processing or no metadata retention.Source for ownStorage (opens in a new tab)
textAndSpeech
Text or uploaded audio scripts. SSML and subtitles are supported by the schema. Speech providers are Microsoft, ElevenLabs, Amazon, Google and Azure OpenAI; default Microsoft. Exact voices, speech models and restrictions depend on the selected provider.Source for textAndSpeech (opens in a new tab)
Audio file size
Uploaded audio URL is limited to 15 MB. Generic schema duration wording names clips and talks only; do not apply those limits to scenes or expressives without evidence.Source for Audio file size (opens in a new tab)
plansMonthly
API monthly plans: Build $18 / 64 credits; Launch $50 / 180, $99 / 360, $149 / 540; Scale $198 / 800, $248 / 1,000, $297 / 1,200. Credits are a shared account allowance, not separate balances for each model.Source for plansMonthly (opens in a new tab)
plansAnnual
Annual API totals: Build 64 credits/month $172.80/year; Launch 180/360/540 credits per month $420/$831.60/$1,251.60 per year; Scale 800/1,000/1,200 credits per month $1,663.20/$2,083.20/$2,494.80 per year. Annual commitment required.Source for plansAnnual (opens in a new tab)
The 1,200-credit Scale card rounds its monthly display to $208; the annual total is $2,494.80.
creditAccounting
One credit covers each started 15-second block of offline video: a 40-second video consumes three credits. Unused monthly credits expire; API and Studio consume the same credit balance.Source for creditAccounting (opens in a new tab)
Included plan allowances
Published API bundles: 64 credits allow up to 16 offline or 32 streaming minutes; 180/360/540 allow 45/90/135 offline or 90/180/270 streaming minutes; 800/1,000/1,200 allow 200/250/300 offline or 400/500/600 streaming minutes. These are alternative uses of a shared balance, not additive free allowances.Source for Included plan allowances (opens in a new tab)
Language coverage
D-ID advertises 120+ languages and accents at platform level. Actual speech coverage depends on the connected TTS provider and selected voice; no separate native D-ID speech-language inventory is asserted.Source for Language coverage (opens in a new tab)
licenseAndWatermark
Build: personal license, D-ID watermark, photo and two video avatars. Launch: commercial license and AI watermark. Scale: commercial license and custom logo. Specific stock/custom-avatar entitlement depends on the plan. Enterprise is quoted separately.Source for licenseAndWatermark (opens in a new tab)
trial
14-day API trial: up to 3 minutes offline or up to 10 minutes streaming within the trial credit capacity, personal license and full-screen watermark. Not unlimited free inference.Source for trial (opens in a new tab)
sentiments
GET /expressives/avatars returns the available avatar and sentiment IDs. Each sentiment has a configured voice; named examples are not a global fixed emotion enum.Source for sentiments (opens in a new tab)
formatConflict
result_url cannot be supplied together with config.result_format, according to the request property. A config description also says formats must match; avoid supplying both.Source for formatConflict (opens in a new tab)
customEnrollment
Multiple recordings capture facial and vocal sentiments. Custom V4 avatars are Enterprise-only according to the model comparison; 24-hour creation is a provider estimate.Source for customEnrollment (opens in a new tab)
performance
Provider claims model latency below 120 ms, conversational latency below 500 ms, and diffusion-pipeline throughput above 200 generated frames/sec. Workload/hardware details are not specified; these measure different stages and are not playback frame rate.Source for performance (opens in a new tab)
runtimeMemory
Provider reports approximately 3.5 GB GPU memory for four concurrent sessions. This is runtime memory, not parameter count or a universal deployment minimum.Source for runtimeMemory (opens in a new tab)
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Video credit consumption
Offline video1
1 credit
started 15 seconds of output
Rate conditions
Published general D-ID video-credit rule; eligible stock/custom avatar access and a funded account are required. Round up each generated video to 15-second blocks. Credit purchase cost depends on subscription and commitment. Not a USD per-minute or LLM token tariff; enterprise terms may differ.
Provider privacy policy describes commercially reasonable deletion of non-persisted API applicative data within 24 hours, Agents within 72 hours, and Insights within 14 days. Persisted assets, knowledge documents and Studio uploads have different lifetimes. Support/operational access is allowed; this is not default zero retention.
The route selects an avatar or presenter ID, not a public immutable weight/checkpoint selector.
Access
POST https://api.d-id.com/v2/agents/{agentId}/sessions; connect using returned LiveKit credentials. interactive sessions use the configured agent pipeline; echo sessions require a secret credential and accept backend-streamed speech.
Regions
No single public-cloud processing region or regional inference guarantee established for this route.
0.5 credit per 15 seconds of generated response video, according to the current Agents pricing FAQ. Subscription, licensed avatars and connected services remain separate.Source for Billing scope (opens in a new tab)
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Agent video credit consumption
Streamed response video1
0.5 credits
15 seconds of generated response
Rate conditions
Current provider Agents FAQ states 0.5 credit for every 15 seconds of response video. Paid plan funding and eligible avatar required. Not wall-clock meeting duration, not the self-hosted license, and not a cash/token tariff.
Provider privacy policy describes commercially reasonable deletion of non-persisted API applicative data within 24 hours, Agents within 72 hours, and Insights within 14 days. Persisted assets, knowledge documents and Studio uploads have different lifetimes. Support/operational access is allowed; this is not default zero retention.
The route selects an avatar or presenter ID, not a public immutable weight/checkpoint selector.
Access
Licensed customer-cloud deployment with GPU rendering and a streaming backend. Enterprise quote and onboarding required; not publicly downloadable open weights. Azure, AWS or GCP infrastructure and Azure/AWS marketplace procurement are described.
Regions
Customer-selected cloud/private network, subject to deployment agreement and any connected external speech/language services.
Pricing terms
Annual platform license including streaming minutes, with additional on-demand minutes. No public dollar tariff; selected AI providers billed separately without stated markup.Source for Pricing terms (opens in a new tab)
scaling
No fixed concurrency cap advertised, but sessions are limited by provisioned infrastructure. GPU sizing is done during onboarding.Source for scaling (opens in a new tab)
latency
Provider advertises below 200 ms avatar-model rendering for self-hosting. This differs from the public V4 technical page’s below-120-ms model claim; deployment/workload equivalence is not established.Source for latency (opens in a new tab)
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Provider says avatar runtime data stays in customer infrastructure and is not sent to D-ID. Chosen external STT, LLM and TTS providers can receive data. This does not establish a blanket no-egress claim for all connected services.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
D-ID
Not verified
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
How it works
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Minimum reusable input
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Lifetime & refresh
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
How to enable it
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Usage fields
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Billing rules
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Limits & data boundaries
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
D-ID
Not verified
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
How it works
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Minimum reusable input
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Lifetime & refresh
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
How to enable it
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Usage fields
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Billing rules
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Limits & data boundaries
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
Customer infrastructure · D-ID license
Not verified
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
How it works
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Minimum reusable input
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Lifetime & refresh
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
How to enable it
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Usage fields
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Billing rules
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
Limits & data boundaries
Not collected
Saved avatars, retained assets and reused result URLs are not a published prompt-cache or discounted inference contract.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
What the measurements say
SyncNet LSE-D
D-ID · provider-reported
7.16distance · lower is better
June 2026 provider study on the V4 technical page: five avatars per company, varied scripts/network conditions, hundreds of videos and tens of thousands of frames. Exact served checkpoint and full artifact set are not disclosed. Not an independent universal ranking; competitor scores are not assigned to guessed competitor checkpoints.
June 2026 provider study on the V4 technical page: five avatars per company, varied scripts/network conditions, hundreds of videos and tens of thousands of frames. Exact served checkpoint and full artifact set are not disclosed. Not an independent universal ranking; competitor scores are not assigned to guessed competitor checkpoints.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
D-ID explicitly lists this generation among its avatar models. The visual rendering technology is proprietary; public SDK source and selectable third-party speech/language providers do not make the visual model weights public.
