D-ID vs HeyGen for Talking-Avatar Ads: Photo Avatars, Lip Sync and the 15-Second Meter

Seven seconds of script, fifteen seconds of plan

A skincare founder takes one well-lit headshot on Monday. By Friday she wants 24 talking-avatar hooks for a serum launch. Eight run 7 seconds, eight run 16 and eight run 31. Short cuts, fast tests.

In D-ID's Studio, every 7-second hook comes off her plan as 15 seconds. That is the rule in D-ID's own pricing FAQ: video length "is rounded up to the nearest 15-second interval." In HeyGen's Studio, the Avatar IV guide counts the other way: "we only charge for the actual seconds of avatar generation." Same photo, same scripts, two different meters.

D-ID and HeyGen both make a still photo talk. Both sell a web studio and an API, and both pitch avatar video to marketers. For a talking-avatar ad, four things decide which one fits: what each builds from a photo, how the lip sync is driven, the unit each one bills in, and how the API is paid for. If you are new to the category, our explainer on how an AI avatar video generator works covers the basics first.

One disclosure. Novoads, which publishes this post, sells AI talking-actor video for ads, so we compete with both. Every D-ID and HeyGen fact below comes from the vendor's own pages, read on September 17, 2026. D-ID's plan and API price tables did not render for the tools we use to verify pages, so this post compares billing units and avatar models, never D-ID dollar prices. Check d-id.com/pricing live before you compare money.

The short answer for ad teams

Read this section if you read nothing else. The rest of the post shows the evidence behind each line.

Choose D-ID when

• You work mostly in the Studio and want one minute balance for the web app and the API.
• The same account also runs visual AI agents: D-ID lists its V2, V3 Pro and V4 avatars as streamable for real-time interactions.
• You want to re-voice real footage: V3 Instant avatars keep the original background and movement.
• Your scripts can be written to land on 15, 30 or 45 seconds.

Choose HeyGen when

• Your hooks are short and uneven in length, since both its Avatar IV Studio meter and its API charge actual seconds.
• You want an API wallet with no subscription behind it.
• The face is not human: a mascot, an animal or a cartoon.
• A winning ad must be translated widely: HeyGen's plans list 175 languages and dialects for translation, and D-ID Video Translate covers up to 29.

Photo avatars: what each one builds from a single image

A photo avatar is the lightest face an ad can have: no footage, no shoot, just a picture of someone who agreed to be animated. Both vendors start there, then offer heavier options built from video.

D-ID: four avatar generations, one of them from a photo

D-ID's avatar page lists four generations, and its comparison table names the input for each: "Single Frontal Image", "1 minute video", "3.5 minute video" and "Multiple videos".

• V2 avatars are "Created from a single image with lightweight rendering." Creation time is immediate, and a custom V2 avatar is available on all plans.
• V3 Instant avatars come from a short video and preserve "the original background and movement while delivering perfectly lip-synced narration." They are for video output only.
• V3 Pro avatars need a longer uploaded video, and a custom one requires Pro and above.
• V4 Expressive avatars are built from several recordings of different emotions, and a custom one is an Enterprise option.

Through the API, D-ID's quickstart puts the photo route in one line: "The Talks endpoint transforms any photo into a speaking avatar." One limit matters for ad creative. D-ID's own AI info page says it is not the right tool when "The content involves non-human faces (cartoon characters, abstract forms, animals): D-ID animates human faces".

HeyGen: one photo, three engines to run it on

HeyGen's avatar page describes the photo route plainly: upload one front-facing photo and the tool "turns that headshot into a talking avatar that speaks your script with natural expression." Its Avatar IV page goes further on inputs, saying the engine "can generate avatars from tilted heads, profiles, and angled poses" and supports "human, anime, and animal avatars in both portrait and full-body formats." For a brand mascot, that is the deciding line. Our Hedra vs HeyGen comparison covers character-led ads in more depth.

The same photo look can run on three engine generations, each metered in credits per minute in HeyGen's app:

1. Avatar III photo look: 7 credits per minute.
2. Avatar IV photo look: 16 credits per minute.
3. Avatar V photo look: 48 credits per minute.

On plans, HeyGen's pricing page lists up to 3 photo avatars on the Free plan and unlimited photo avatars on Creator and above.

Consent comes before the first render

Both vendors put consent on the page. D-ID says personal avatars "can only be created with explicit consent," and its API quickstart for V3 Instant avatars requires "a consent verification process to prevent misuse" before training. HeyGen says "Custom avatars require verified consent from the person being recreated."

For an ad, the practical rule is simple. The face in the photo belongs to someone who signed off: the founder, an employee, or a creator under contract. For picking and shooting that source image, see our guide to making an AI avatar from a photo. For how Meta, Google and TikTok treat synthetic people in ads, see the rules on labeling AI-generated ads.

Lip sync: a script, your own audio, or a translation

A talking-avatar ad fails fastest on the mouth. Both vendors drive lip sync from a script or an uploaded recording, and both can re-sync an existing video to a new language.

Driving the mouth from text or a recording

• D-ID: its API page says you give the presenter a voice "by choosing from hundreds of available text-to-speech options or uploading an audio recording of your own." D-ID says its avatars can narrate "in more than 120 languages and accents."
• HeyGen: its Avatar IV page says you can "Type your script directly into the editor or upload an audio file. You can also record your own voice." Its pricing page lists 175+ languages and dialects on Creator.

Bring your own recording when the read matters, such as a founder talking about a product she makes. Both vendors accept one, so the voice never has to be the compromise.

Re-syncing a finished video in another language

• D-ID Video Translate "Adapts lip movements to match the new language," clones the speaker's voice, translates into "as many as 29 languages," and is "Accessible through the self-service studio and API." You can proofread the translated text before generating.
• HeyGen charges lip-synced translation at 6 credits per minute in speed mode and 10 in precision mode inside its app. Its API lists lip-synced translation at $0.81 per minute in speed mode and $1.50 in precision mode, and a separate lipsync audio mode where "you can upload your own audio to be lipsynced to the video" at $0.84 and $1.74 per minute.

What to check in a lip-sync test

Run one script through both before you commit to either:

1. A line heavy with p, b and m sounds, where a closed mouth is easy to miss.
2. Your product name, said twice, fast.
3. The call to action at speaking pace, not read-aloud pace.
4. A three-quarter face angle. D-ID's translate page says lip sync "is most accurate when the face is clearly visible."

The billing unit: 15-second blocks or actual seconds

Plan prices get compared. The meter underneath decides what a batch of short hooks consumes, and here the two vendors genuinely differ.

D-ID: plan minutes in 15-second steps

D-ID's Studio pricing FAQ says the duration of each video "is deducted from the total minutes available in your plan," and adds the rule that shapes every short ad: "The length of the video is rounded up to the nearest 15-second interval. For example, a 1-minute and 10-second video will consume 1 minute and 15 seconds."

Two more lines from the same FAQ complete the picture. Unused minutes do not roll over: "Minutes do not accumulate, they are renewed every month and unused minutes become void." And API usage shares the pool: "Minutes used via the API are deducted from the same balance as your web version of D-ID." The rounding rule is stated for plan minutes in that FAQ, so if your batch runs through the API, confirm with D-ID how those renders are rounded before you budget. Our roundup of D-ID alternatives sets this meter against other avatar tools.

HeyGen: credits per minute, charged by the second

Inside HeyGen's app, avatar video spends credits at the per-minute rates above. The Avatar IV guide adds how Studio counts: "we only charge for the actual seconds of avatar generation," and its example says "a 30s video might only use 20 or 25 seconds of credits, depending on pauses or silent moments."

Creator costs $29 a month, or $24 paid annually, and includes 600 credits. HeyGen's pricing FAQ says monthly subscribers' unused credits roll over for one more month, and its credits article says Creator does not support one-time credit purchases, so running short means upgrading, a limit our HeyGen vs InVideo comparison also flags.

What a rounding meter does to a hook

Under a 15-second rule, the waste depends on where each script ends:

• A 7-second hook consumes 15 seconds, more than double.
• A 16-second hook consumes 30.
• A 31-second hook consumes 45.
• A 45-second hook consumes exactly 45.

A rule of thumb: on a 15-second meter, a script that ends on 15, 30 or 45 seconds wastes nothing, and one second over pays for fourteen more.

The API: one shared balance or a separate wallet

If ads are generated from a spreadsheet or a product feed, the API is the product. The two vendors package it in opposite ways.

D-ID: the web app's minutes, through an API key

D-ID's pricing FAQ says to log in, open your Account page, and "generate your API key." From there:

• API minutes come out of the same balance as the Studio.
• The Talks endpoint returns a talking video from a photo, and the quickstart says "The video typically takes 10-30 seconds to process."
• "The video URL is valid for 24 hours," so store each render as soon as it lands.
• D-ID's avatar table lists V2, V3 Pro and V4 avatars as streamable for real-time interactions, and V3 Instant as video only.

D-ID also publishes an API pricing page, but its price table did not render for our fetcher, so we do not quote it.

HeyGen: pay-as-you-go credits with no plan

HeyGen's API pricing article, last updated on September 16, 2026, says: "No plan, no subscription needed." You buy credits in US dollars, separately from Creator, Pro or Business. The table prices avatar video "per 1 minute, charged by actual seconds generated":

• Avatar III photo avatar: $0.99 per minute.
• Avatar IV photo avatar: $2.31 per minute.
• Avatar IV digital twin: $4.83 per minute.
• Avatar V digital twin: $7.20 per minute.
• Creating a photo avatar: $1.32 per call.

The fine print matters for a pipeline. Pay-as-you-go credits expire after 12 months, the API processes 10 videos concurrently, HeyGen stopped offering free API credits in February 2026, and creating a custom digital twin through the API is Enterprise-only. Streaming is priced on its own: HeyGen says "LiveAvatar plans and pricing are separate from API plans." HeyGen's older Avatar IV guide still quotes a different API figure, so budget from the dated API article, not from older pages.

Two wallets or one

D-ID's single pool is simpler to watch, but API renders compete with the team's Studio work for the same monthly minutes, and whatever is left voids at renewal. HeyGen splits plan credits from API credits, which means two balances, but API spend never eats the team's plan credits. HeyGen also lets integrations such as Zapier, Make and n8n spend plan credits instead of API credits, which suits a no-code pipeline.

A worked example: 24 hooks for one serum launch

Back to the founder from the opening. One photo avatar, 24 scripts: eight at 7 seconds, eight at 16 and eight at 31. That is 432 seconds, or 7.2 minutes, of footage.

On D-ID's Studio meter

Each 7-second hook consumes 15 seconds, each 16-second hook 30, and each 31-second hook 45. The batch consumes 120 + 240 + 360 = 720 seconds, or 12 plan minutes, about 67% more than the footage.

On HeyGen's meters

Each figure below is minutes of video times the published rate, before any retakes.

• Avatar IV photo avatar API: 7.2 × $2.31 = $16.63, plus $1.32 if you create the photo avatar through the API.
• Avatar III photo avatar API: 7.2 × $0.99 = $7.13.
• Studio, Avatar IV photo look: 7.2 × 16 = about 115 credits of Creator's 600, or fewer where the scripts pause.

Trim the scripts to the boundary

Now cut every 16-second script to 15 seconds and every 31-second script to 30. Footage drops to 416 seconds.

Version	Footage	D-ID Studio	HeyGen API, Avatar IV photo
As written	432 s	12 plan minutes	$16.63
Trimmed	416 s	8 plan minutes	$16.02

On D-ID, trimming one second from 16 of the 24 scripts removes a third of the plan minutes. On HeyGen, the same edit saves about 60 cents. On a 15-second meter, trimming is a pricing decision. On a per-second meter, it is only an editing decision. How many hooks a test actually needs is its own question, worked through in our guide to how many ad creatives you need.

Which one fits your ad program

The worked example prices the batch. Fit depends on what else the account has to do.

Where D-ID is the better-shaped tool

• Ads and visual agents on one account. Streamable V2, V3 Pro and V4 avatars can front a support agent and a video ad with the same face.
• A small team in the Studio. One balance for web and API is one number to watch.
• Re-voicing real footage. V3 Instant keeps the original background and movement, and Video Translate clones the speaker's voice.
• Writers who can hit a boundary. Scripts built to 15, 30 or 45 seconds lose nothing to the rounding.

Where HeyGen is the better-shaped tool

• Short, uneven hooks. Actual-second billing in the Avatar IV Studio meter and in the API.
• An API without a subscription, with published per-engine rates and credits that last 12 months.
• Non-human faces, from mascots to animals.
• A bigger cast. Its avatar page lists 1,100+ AI presenters ready to use, and bulk generation "turns a spreadsheet of scripts into finished videos in one run."

If HeyGen's presenter style is not what your ads need, our list of the best HeyGen alternatives for UGC ads covers tools built for handheld creative.

Run the same test on both

1. Use one consented photo and one recorded audio file on both, so the voice is not the variable.
2. Render a 7-second hook and a 31-second spot on each.
3. Judge lip sync with the checklist above, then compare what each account deducted.
4. Put the survivors into a small paid test and read it properly, as in our guide to ad creative testing.

How Novoads solves the product-in-hand talking ad

Novoads is an AI UGC video-ad generator: upload a product image or write a script, pick an AI actor, and get an ad-ready video in 9:16, 1:1 or 16:9. It has a photo route too. Its Create Actor dialog offers an upload option to "Transform a picture into a talking actor," so a photo of someone holding your product becomes a custom actor that shows it in hand. Discover's product templates take another route, placing your uploaded product in a template creator's hand. Actors come with lip sync and voices in 31 languages.

The concessions are real. A Novoads talking-actor video costs 10 credits per started minute, a coarser block than D-ID's 15 seconds or HeyGen's actual seconds. Rendered one by one, the 24 hooks above would cost 240 credits, so on Novoads it pays to script several short hooks into one render of up to a minute and cut them apart in your editor. Novoads markets 100+ AI actors, far fewer than HeyGen's library, and it does not offer real-time agents like D-ID's.

Novoads starts at $49/month (Starter, 50 credits per month), and all plans are published on /pricing. If your ad is a person talking about a product, try Novoads beside your D-ID or HeyGen renders and compare the output.

Write to the meter you pay for

The founder's hooks said the same thing in both versions of the worked example. Only a second of script changed, and on one of the two meters that second moved the bill by a third. A talking-avatar ad is priced twice: once on the rate card and again at the meter. D-ID rewards scripts written to 15-second boundaries. HeyGen rewards scripts written to exactly the length the hook needs. Decide which kind of scripts your team writes, then pick the tool whose meter agrees with them.