HeyGen wins at 9.1/10: 200 personalized videos a month for $1,440 a year via API and Zapier, n8n or Make. One rival costs about 20x more for the same job.
Wiring an AI video API into Zapier, n8n or Make takes an afternoon. Keeping it running for a year is the hard part: the batch that stops halfway because the credit pool emptied, the 429s when fifty form fills land at once, the minutes that expired last Tuesday.
So I priced eight of the best AI video generation APIs on one identical automated job: 200 personalized 60-second presenter videos a month, triggered from a CRM and returned by webhook. Every tool was scored on the same six weighted criteria, from published rate cards, limits and terms.
HeyGen won at 9.1 out of 10. It runs the job for $1,440 a year on its API, and it is the only presenter API here with listed connectors for Zapier, Make and n8n plus an MCP route that needs no API key.
The short answer
- Best overall: HeyGen, 9.1/10. Pay-as-you-go API from $5, 200 personalized digital-twin videos for $120 a month, listed Zapier, Make and n8n integrations.
- Best presenter realism per dollar: Tavus, 8.2/10. Builder at $59 a month plus $1 per video-generation minute, $2,928 a year for the job.
- Cheapest fixed plan: D-ID API Scale, 8.1/10. Up to 200 video minutes for $1,663 a year billed annually.
- Best no-code render layer: Creatomate, 7.8/10. Native Zapier, Make and n8n connectors for templates, but no lip-synced presenter.
- Overpriced for this job: Synthesia, 6.7/10. Its self-serve API can't clear 200 minutes, and the one public Enterprise contract runs $29,880 a year.
- Method: one priced job, one evidence checklist, six weighted criteria, prices verified November 2026.
The Short Version
- HeyGen, 9.1 Best overall. Avatar III digital-twin videos at $0.60 a minute via API, with listed Zapier, Make and n8n integrations and webhook delivery. For sales, marketing and product teams automating personalized video.
- Tavus, 8.2 Best replica realism per dollar. One API covers recorded video and real-time conversation, with uncapped pay-as-you-go overage. For products that need one replica across recorded and live video.
- D-ID, 8.1 Cheapest fixed monthly plan. API Scale covers up to 200 minutes, with streaming minutes at double the video allowance. For developers who want a flat bill and portrait-style talking heads.
- Synthesia, 6.7 Best enterprise governance. Template control and enterprise compliance, but API access starts on Creator and volume needs an Enterprise contract. For L&D teams already on Synthesia Enterprise.
- Creatomate, 7.8 Best no-code render layer. Native Zapier, Make and n8n connectors for template-driven video, priced by pixels and frames. For teams assembling captions, B-roll and branding around a presenter from another API.
- fal.ai, 7.4 Best multi-model clip API. One key for hundreds of hosted video models, billed only for successful outputs. For developers adding generated B-roll to an automated pipeline.
- Runway API, 7.0 Best per-task cost visibility. Its Task Cost API returns estimated and final cost per job. For creative pipelines built on Runway's own models.
- Google Veo 3.1 (Gemini API), 6.6 Best for Google Cloud shops. Veo 3.1 with native audio, Lite and Fast tiers for cheaper drafts. For teams already billing through Google.
Scores and prices at a glance
Prices verified November 3, 2026. "Job" means 200 personalized 60-second presenter videos a month, 1080p, commercial license, webhook delivery.
Which AI Video API Should You Pick?
Why I Judge Tools This Way
I judge video APIs against one question: if a form fill or a CRM stage change fires at 2 a.m., does a finished personalized video land in the right inbox without anyone touching it? A clean demo call answers almost none of that.
So the evidence here is the boring part of every API: what it costs per finished minute, what happens when credits run out, how many jobs run at once, whether it reports failure, and whether Zapier, n8n or Make can reach it without custom code.
How I Compared
Every tool got the same priced job and the same evidence checklist: 200 personalized 60-second presenter videos a month at 1080p, commercial license, triggered from an automation platform and returned by webhook, priced on the cheapest route each vendor sells that clears it.
The checklist covered each vendor's API pricing page, plan limits, credit expiry and top-up rules, concurrency limits, webhook and failure reporting, listed integrations for Zapier, Make and n8n, and the license terms. Prices were read between October 9 and November 3, 2026.
Fit for the automation job (25%)
I scored whether the API can produce a script-faithful 60-second presenter video with variables, and whether Zapier, Make and n8n can call it through listed integrations or only through generic HTTP steps. Webhooks, batch endpoints and templates add points.
Real cost of the priced job (25%)
I computed the 12-month cost of 200 one-minute videos on the cheapest route that clears the job, including expiry, top-up rules and overage. For APIs that can't make a presenter, this score reflects their per-second price for the B-roll they'd add to the same workflow, and the fit score carries the penalty.
Output quality (15%)
I scored presenter realism on the engine actually priced, not the vendor's best engine. A tool priced on a cheaper engine is scored on that engine.
Reliability at volume (15%)
I scored published concurrency limits, rate-limit behavior, failure events, and whether a batch stops silently when the balance runs dry.
Time to first working workflow (10%)
I scored the path from signup to a running automation: trial access, listed connectors, an MCP route, and documentation.
Exit and rights (10%)
I read the license terms, watermark rules, credit expiry and data-use statements. Silence is scored as a finding.
Rubric anchors: 9 to 10 means nothing in the category failed and the tool did something no other tool here does. 7 to 8 means it works with one annoyance. 5 to 6 means it needs a workaround. 3 to 4 means cost or rights terms change the buy. 0 to 2 means it can't do the job.
The Two Factors That Decided This
The first decider was workflow fit. Four of the eight APIs here can't produce a 60-second presenter reading a personalized script at all; their models return 2- to 10-second clips. Of the four that can, only HeyGen lists native integrations for Zapier, Make and n8n on its own integrations page, which is why fit carries 25%.
The second decider was the hidden cost mechanic, and it hit Synthesia hardest. Synthesia's Creator plan includes API access but no self-serve top-ups, credits don't roll over, and generation stops when the balance can't cover a job. At 75 credits per API minute, Creator's 3,600 monthly credits cover 48 minutes, a quarter of the job, which is why cost also carries 25%.
Tier 1: Presenter APIs that clear the job
These four produce a script-faithful presenter video from an API call. What this tier cannot do: none of them generates cinematic B-roll at the per-second prices of Tier 3, and none assembles complex branded templates as cheaply as Creatomate.
1. HeyGen: Best Overall AI Video API for Automation
- Score: 9.1/10
- Entry price: API pay-as-you-go from $5, no subscription
- Cost for 200 one-minute personalized videos a month: $120/mo, $1,440/year, $0.60 per finished minute (Avatar III digital twin)
- Metering: Per second generated; Avatar III twin $0.60/min, Avatar IV twin $4.83/min, Avatar V twin $7.20/min
- Volume cliff: None on cost; 10 concurrent jobs on self-serve, 20 on Enterprise
- Free tier: No free API credits since February 2026; MCP works on every plan, including Free
- Priced on: HeyGen API v3, Avatar III digital twin, November 2026
HeyGen's API bills per second of finished video from a prepaid wallet, with no subscription and no monthly minimum beyond the $5 starting balance. A sales rep's digital twin rendered on Avatar III costs $0.60 a minute, so 200 personalized one-minute videos come to $120 a month.
What makes it the automation pick is the plumbing. HeyGen's integrations page lists Zapier, Make, n8n, Pabbly and viaSocket, the API runs batches and webhooks, and the AI video generator is also reachable through HeyGen MCP with OAuth sign-in and no API key.
What it wins: workflow fit, with listed Zapier, Make and n8n integrations plus an MCP route, against generic HTTP calls for the other presenter APIs.
How it actually works: you build a template with variables in AI Studio, then fire one call per contact with name and company filled in. The API processes 10 videos at once on self-serve, returns a 429 with a Retry-After header above that, and Enterprise can opt into burst capacity billed at 1.5x.
Where it falls short: HeyGen lost output quality to Tavus on the engine I priced. Avatar III is HeyGen's cheapest engine; Avatar IV and V close the gap at $4.83 and $7.20 a minute. Through the API, Avatar IV and V don't render native 4K; a 4K request returns an upscaled 1080p avatar.
Client example: Videoimagem produced 50,000+ personalized videos for AB InBev on HeyGen, with up to 3x engagement (story).
Pros:
- $0.60 per minute for a digital twin on Avatar III, the lowest presenter rate here
- Listed integrations for Zapier, Make, n8n, Pabbly and viaSocket
- MCP with OAuth works on every plan, with no API key to manage
- Scoped API keys limit a leaked key to the endpoints it needs
- One API covers avatars, Video Agent, translation at $0.81/min and HeyGen Video 1 clips
Cons:
- API funds are a separate wallet from web-plan credits, and they expire after 12 months
- Self-serve concurrency is 10 jobs, so bursts above that queue on 429 retries
Category scores: Fit 9.5 | Cost 9.3 | Quality 8.6 | Reliability 8.8 | Time to workflow 9.0 | Exit and rights 8.8
2. Tavus: Best Replica Realism Per Dollar
- Score: 8.2/10
- Entry price: Starter $22/mo; Builder $59/mo
- Cost for 200 one-minute personalized videos a month: $244/mo on Builder with overage, $2,928/year, $1.22 per finished minute
- Metering: Included video-generation minutes, then $1.00 (Builder), $0.90 (Growth) or $0.80 (Business) per minute
- Volume cliff: None; overage bills without a cap
- Free tier: Free plan with conversation and video-generation minutes
- Priced on: Builder plan, June to October 2026 captures
Tavus prices recorded video as a minute allowance plus uncapped overage. Builder includes 15 video-generation minutes, so the priced job adds 185 overage minutes at $1 each.
The draw is the replica. A replica trained once stars in recorded videos and in real-time conversations through the same API, which matters for products that mix outbound video with a live agent.
What it wins: output quality on the priced route, with replica realism scored above HeyGen's Avatar III.
How it actually works: you train a replica, then call video generation with a script per contact. Tavus also exposes personas, conversations and guardrails through the same REST API.
Where it falls short: the job costs about twice HeyGen's yearly total, and I found no listed native n8n or Make module, so most no-code builds go through HTTP steps.
Client example: no verifiable client result with a time or cost delta was found.
Pros:
- One replica covers recorded video and real-time conversation
- Overage bills instead of blocking, so batches don't stop mid-run
- Overage drops to $0.80 a minute on Business
- Image-to-replica training from a single photo, added May 2026
Cons:
- The job costs $2,928 a year, 2x HeyGen's total
- Plan packaging changed several times in 2026, so budget against the live page
Category scores: Fit 8.2 | Cost 8.4 | Quality 8.8 | Reliability 8.0 | Time to workflow 7.8 | Exit and rights 7.8
3. D-ID: Cheapest Fixed Monthly Plan
- Score: 8.1/10
- Entry price: API Build 14.40 billed annually), personal-use license
- Cost for 200 one-minute personalized videos a month: $138.60/mo on Scale billed annually, $1,663/year, $0.69 per finished minute
- Metering: Video minutes per plan, with streaming minutes at double the video allowance
- Volume cliff: Scale Small caps at about 200 video minutes a month
- Free tier: 14-day trial
- Priced on: API Scale Small, October 2026
D-ID sells API minutes as fixed monthly packs. Scale Small covers up to 200 video minutes for $198 a month, or $138.60 billed annually, which clears the job exactly.
That exactness is the risk. Minutes don't carry over, and a 201st video waits for next month or a bigger pack.
What it wins: the most predictable flat bill among presenter APIs.
How it actually works: you send a photo or avatar ID with a script, and D-ID returns a talking-head clip. Commercial licensing starts at Launch; the Build tier is personal use with a D-ID watermark.
Where it falls short: output is portrait-style talking heads, scored below HeyGen and Tavus on realism, and Launch-tier videos carry an AI watermark.
Client example: no verifiable client result with a time or cost delta was found.
Pros:
- Flat annual price: $1,663 for up to 200 minutes a month
- Streaming minutes at double the video allowance on the same plan
- Cloned voices included from Launch and Scale
- Low entry at $18 a month for prototyping
Cons:
- Hard cap at 200 minutes on Scale Small, with unused minutes expiring monthly
- Build is personal-use only and watermarked, so prototypes can't ship
Category scores: Fit 8.0 | Cost 9.1 | Quality 7.4 | Reliability 7.6 | Time to workflow 8.4 | Exit and rights 7.6
4. Synthesia: Best Enterprise Governance
- Score: 6.7/10
- Entry price: Creator 59 billed annually), the first plan with API access
- Cost for 200 one-minute personalized videos a month: cannot clear on self-serve; one public Enterprise contract lists $29,880/year
- Metering: 75 credits per API minute, 90 per bulk-personalization minute, 120 per standard minute
- Volume cliff: Creator's 3,600 monthly credits cover 48 API minutes, with no self-serve top-ups
- Free tier: Free plan without API access or downloads
- Priced on: Creator and Enterprise, August 2026 plan documentation
Synthesia's API sits on the Creator plan and draws from the same credit pool as the editor. The pool resets monthly, doesn't roll over, and stops generation when it can't cover a job.
For 200 minutes a month, that means Enterprise. The one public Enterprise reference is an AWS Marketplace contract: $29,880 for three users over 12 months with unlimited video minutes.
What it wins: enterprise governance, with SSO, SCORM and brand controls at the Enterprise tier.
How it actually works: API calls render template-based avatar videos, and bulk personalization runs at 90 credits a minute. Administrators get alerts at 75%, 90% and 100% of advanced-feature credits.
Where it falls short: cost and continuity. A batch that exceeds Creator's pool stops rather than bills, which is the failure an automation can't absorb.
Client example: no verifiable public client result with a time or cost delta was found for API automation.
Pros:
- Enterprise tier includes SSO, SCORM and brand controls
- API generation at 75 credits a minute, below the 120-credit editor rate
- Usage alerts at 75%, 90% and 100% of advanced credits
- Strong enterprise L&D incumbency and template control
Cons:
- Creator's API covers 48 minutes a month with no top-ups, so the job needs Enterprise
- Generation stops when the balance can't cover a job, which breaks unattended batches
Category scores: Fit 7.6 | Cost 3.4 | Quality 8.6 | Reliability 7.4 | Time to workflow 7.6 | Exit and rights 8.2
Tier 2: Template render layer
Creatomate assembles video from templates, captions, images and stock, and it is the easiest of these to drive from no-code tools. What this tier cannot do: render a lip-synced presenter reading a personalized script. Pair it with a Tier 1 API for that.
5. Creatomate: Best No-Code Render Layer
- Score: 7.8/10
- Entry price: Essential 45 billed annually) for 2,000 credits
- Cost for 200 one-minute personalized videos a month: cannot clear alone; the render layer costs $129/mo on Growth, $1,548/year
- Metering: Credits scale with pixels and frame rate; about 37 credits per minute at 1080p30, about 14 at 720p
- Volume cliff: Essential's 2,000 credits cover about 54 minutes at 1080p
- Free tier: 50 one-time trial credits
- Priced on: Growth plan, October 21, 2026
Creatomate is the glue tier. Native connectors for Zapier, Make and n8n let a spreadsheet row or form fill populate a branded template without writing a REST call.
It doesn't render a talking presenter. Teams pair it with a presenter API, then use Creatomate for captions, lower thirds and B-roll around that clip.
What it wins: the deepest no-code connector support in this list, with native Zapier, Make and n8n integrations.
How it actually works: you design a template in its editor, expose fields as variables, and trigger renders from a connector or the REST API. Automatic subtitles cost 10 credits a minute only when that exceeds the render cost.
Where it falls short: pricing scales with pixels, so 1080p costs about 2.6x 720p per minute, and no lip-synced presenter exists in the product.
Client example: no verifiable client result with a time or cost delta was found.
Pros:
- Native connectors for Zapier, Make and n8n
- Template variables map cleanly to CRM or spreadsheet fields
- Subtitles are effectively free at 720p and above
- Webhook delivery on finished renders
Cons:
- No lip-synced presenter, so personalized talking video needs a second vendor
- Credits scale with pixels, making 1080p and 4K far pricier than 720p
Category scores: Fit 7.8 | Cost 8.6 | Quality 6.0 | Reliability 8.0 | Time to workflow 8.2 | Exit and rights 7.6
Tier 3: Generative clip APIs
These three generate new footage from text or images, billed per second. What this tier cannot do: deliver a script-faithful 60-second presenter; clips run 2 to 10 seconds per call. They belong in the workflow as the B-roll step.
6. fal.ai: Best Multi-Model Clip API
- Score: 7.4/10
- Entry price: Pay per output, no subscription
- Cost for 200 one-minute personalized videos a month: cannot clear the job; B-roll priced per second by model
- Metering: Per second or per video, by model (Wan 2.5 $0.05/sec, Kling 2.5 Turbo Pro $0.07/sec, Veo 3 $0.40/sec)
- Volume cliff: None published for model APIs
- Free tier: Promotional credits for new accounts, expiring in 90 days
- Priced on: fal model API pricing, August to October 2026 captures
fal puts hundreds of hosted image and video models behind one key and one queue. You pay only for successful outputs; server errors and queue time are free.
For an automation, that means one integration can swap models as prices move. It doesn't hold your avatar, voice or brand template, so the presenter step lives elsewhere.
What it wins: model choice, with dozens of video models behind one API key.
How it actually works: you submit a job to the queue, then receive the result by webhook or poll. Purchased credits expire 365 days from purchase.
Where it falls short: no persistent presenter, template or brand layer, so personalization logic is yours to build.
Client example: no verifiable client result with a time or cost delta was found.
Pros:
- One key for many video models, from $0.05 to $0.40 per second
- Billed only for successful outputs
- Queue and webhooks built for bursts
- New models appear within days of release
Cons:
- No avatar, voice or template layer for personalized presenter video
- Per-model pricing makes the monthly bill depend on which model each job picks
Category scores: Fit 5.8 | Cost 8.0 | Quality 8.4 | Reliability 8.2 | Time to workflow 7.6 | Exit and rights 7.4
7. Runway API: Best Per-Task Cost Visibility
- Score: 7.0/10
- Entry price: API credits at $0.01 each
- Cost for 200 one-minute personalized videos a month: cannot clear the job; Gen-4.5 B-roll at $0.12/sec
- Metering: 12 credits per second for Gen-4.5, 5 for Gen-4 Turbo
- Volume cliff: None on the API wallet
- Free tier: None on the API
- Priced on: Runway API rate card, 2026
Runway's API is the cleanest about money. Its Task Cost API, added July 30, 2026, returns each task's estimated cost before the run and its final cost after.
Clips run 2 to 10 seconds, so Runway fits the B-roll or scene step of a workflow, not the presenter step.
What it wins: per-task cost reporting, with estimated and final cost returned per job.
How it actually works: you call Gen-4.5 at $0.12 a second or Gen-4 Turbo at $0.05 a second, from a wallet separate from Runway's web subscriptions.
Where it falls short: subscription credits don't transfer to the API, so teams using both pay twice, and Gen-4.5 renders natively at 720p.
Client example: no verifiable client result with a time or cost delta was found.
Pros:
- Task Cost API returns estimated and final cost per job
- Gen-4 Turbo at $0.05 a second for cheap drafts
- Act-Two performance capture at 5 credits a second
- Same models as Runway's web app
Cons:
- Web subscription credits don't work on the API
- No presenter or template layer for personalized video
Category scores: Fit 5.4 | Cost 7.4 | Quality 7.6 | Reliability 8.0 | Time to workflow 7.4 | Exit and rights 7.6
8. Google Veo 3.1 (Gemini API): Best for Google Cloud Shops
- Score: 6.6/10
- Entry price: Pay per second on the Gemini API
- Cost for 200 one-minute personalized videos a month: cannot clear the job; Veo 3.1 B-roll at $0.40/sec at 1080p with audio
- Metering: Per second; an 8-second Lite clip at 720p video-only costs $0.24
- Volume cliff: None published
- Free tier: Daily Flow credits in the app, not on the API
- Priced on: Gemini API rates via Artificial Analysis and DIYAI, November 2026
Veo 3.1 on the Gemini API suits teams whose cloud bill already runs through Google. Lite and Fast tiers make drafts cheap, and the full model generates native audio in the same pass.
Each call returns up to 8 seconds, so like Runway and fal it serves the B-roll step.
What it wins: native audio in one pass, with Lite pricing at $0.24 per 8-second 720p clip for drafts.
How it actually works: you call the Gemini API with a prompt and optional first and last frames, then poll for the result. Pricing varies by tier, resolution and audio.
Where it falls short: the full Veo 3.1 rate of $0.40 a second is the highest per-second price in Tier 3, and no presenter layer exists.
Client example: no verifiable client result with a time or cost delta was found.
Pros:
- Native audio generated with the video
- Lite and Fast tiers for cheap drafting
- First and last frame control
- Billing through an existing Google relationship
Cons:
- $0.40 a second for full Veo 3.1 at 1080p with audio
- Clips cap at 8 seconds per generation
Category scores: Fit 5.0 | Cost 6.0 | Quality 8.0 | Reliability 7.8 | Time to workflow 7.2 | Exit and rights 7.6
What 200 Automated Personalized Videos a Month Actually Costs
The job: 200 personalized 60-second presenter videos a month, 1080p, commercial license, triggered from Zapier, n8n or Make and returned by webhook. Prices verified November 3, 2026.
Check one row: HeyGen's API charges $0.60 per minute for a digital twin on Avatar III, billed by actual seconds generated. 200 videos × 1 minute × $0.60 = $120 a month, × 12 = $1,440.
Check a second row: Tavus Builder costs $59 a month and includes 15 video-generation minutes. The other 185 minutes bill at $1.00 each, so $59 + $185 = $244 a month, × 12 = $2,928.
Where the cheap options cross over
On the engine I priced, HeyGen stays cheapest at every volume up to D-ID's 200-minute cap: $0.60 a minute against D-ID's effective $0.69. Past 200 minutes, D-ID needs a bigger pack, while HeyGen keeps billing per second.
The line moves if you want HeyGen's more expressive engines. On Avatar IV at $4.83 a minute, Tavus Builder becomes cheaper above about 12 minutes a month, so teams that need top realism at volume should price both.
Which tool is overpriced for this job
Synthesia's one public Enterprise contract runs $29,880 a year against HeyGen's $1,440 for the same 200 monthly minutes, about 20.8x the spend, and it scores 6.7 against 9.1. What the difference buys is enterprise governance: SSO, SCORM, brand controls and a long L&D track record.
That contract covered three users and unlimited minutes, so a negotiated quote may land elsewhere. The self-serve gap stays: Creator's API clears 48 minutes, not 200.
What leaving costs
HeyGen's API funds expire 12 months after purchase, and HeyGen states customer data isn't used to train its models. D-ID minutes expire monthly, Synthesia credits reset each period, and fal credits expire after 365 days. Tavus's published pages I read were silent on what happens to a trained replica after cancellation.
AI Video Generation API Comparison: Best AI Video Generation APIs by Spec
"HTTP route" means I found no listed native connector in the sources I read; every API here is callable from Zapier Webhooks, Make's HTTP module and n8n's HTTP Request node.
Which One Is For You
By the work you are doing
- Personalized sales or onboarding video from CRM events: HeyGen, at $0.60 a minute on a digital twin.
- One replica that also talks live on your site: Tavus, with recorded and real-time video on one API.
- Branded captions and B-roll around an existing presenter: Creatomate, through its native connectors.
- Generated B-roll inside a content pipeline: fal.ai, with many models behind one key.
By budget
- Free: HeyGen MCP on the free web plan, for testing prompts from Claude or ChatGPT without an API key.
- Under $150 a month: HeyGen, at $120 for the full 200-video job.
- Over $2,000 a month: Synthesia Enterprise, when governance requirements outrank cost.
By learning-curve tolerance
- I want output today: HeyGen, through MCP or its listed Zapier integration.
- I will spend a weekend: D-ID, wiring its REST API through n8n HTTP nodes.
- I have a team to train: Synthesia Enterprise, with SSO and brand controls.
Too New To Rank
Synthesia Syren, an agent that builds avatar videos from a description or document, launched in October 2026. It enters the ranking once its API pricing is public and has a billing cycle behind it.
Synthesia Sessions and its Interactive Avatar API (July and October 2026) compete with real-time avatar platforms rather than recorded-video APIs, so they'd belong in a separate live-avatar ranking.
Utopai X ranks near the top of Artificial Analysis' image-to-video board but lists no API, so it can't be wired into an automation yet. I'll retest all three in February 2027.
Workflow Walkthrough: CRM Row to Personalized Video to Inbox
This is the pattern behind most automated personalized video, shown with HeyGen. The same five steps apply in Zapier, Make or n8n.
- Build one template. In HeyGen's AI Studio, record or pick the presenter, write the script, and mark the personalized fields as variables: first name, company, and one custom line.
- Pick the trigger. A new CRM deal stage, a form submission or a spreadsheet row starts the workflow in Zapier, Make or n8n.
- Call the API. Pass the row's fields into the template's variables. For volume, send batches and respect the 10-job concurrency limit by retrying on 429 after the Retry-After interval.
- Wait for the webhook. Don't poll in a loop. Register a webhook so the finished video URL comes back to the workflow when rendering completes.
- Deliver and log. Send the link by email or CRM task, then write the video URL back to the contact record.
For teams that would rather not hold an API key, HeyGen MCP connects Claude, Cursor or ChatGPT to the same account with OAuth. It suits ad hoc batches; scheduled volume belongs on the API.
The trap to avoid is mixing budgets. HeyGen's web-plan credits don't fund Zapier, Make or n8n runs; those draw from API credits, so top up the API wallet before the campaign, not during it.
Frequently Asked Questions
What is the best AI video generation API for automation in 2026?
HeyGen, at 9.1/10. It renders 200 one-minute personalized digital-twin videos for $1,440 a year and lists integrations for Zapier, Make and n8n. Tavus is second at 8.2, with stronger replica realism on the engines I priced.
Can I use HeyGen with Zapier, Make or n8n?
Yes. HeyGen's integrations page lists Zapier, Make, n8n, Pabbly and viaSocket, and the API returns results by webhook. Those runs bill API credits, not web-plan credits, so fund the API wallet separately from your subscription.
Is there a free AI video generation API?
Not for production use. HeyGen ended free API credits in February 2026 and starts at $5, D-ID offers a 14-day trial, and Creatomate gives 50 trial credits. HeyGen MCP does work on the free web plan for testing.
Is the Sora API still available for video automation?
No. OpenAI shut the Sora app in April 2026, and the Sora API was scheduled to sunset in September 2026. Build new pipelines on HeyGen, Runway, Veo 3.1 or a fal-hosted model instead.
Is HeyGen or D-ID better for personalized sales videos at scale?
HeyGen, at 9.1 against D-ID's 8.1. HeyGen costs $1,440 a year for 200 one-minute twin videos against D-ID's $1,663, and it keeps billing past 200 minutes. D-ID wins if you also need streaming minutes on the same plan.
What is the most common mistake when automating AI video?
Polling and bursting. Firing more jobs than the concurrency limit returns 429 errors, and polling in a loop burns automation tasks. Use webhooks, honor Retry-After, and avoid plans like Synthesia Creator where generation stops when credits run out.
How much does an AI video API cost per minute?
Between $0.60 and $7.20 for presenter video. HeyGen runs $0.60 on an Avatar III twin, $4.83 on Avatar IV and $7.20 on Avatar V; D-ID's Scale pack works out to $0.69, and Tavus overage runs $0.80 to $1.00.
Do I need to code to generate AI video from my CRM?
No. HeyGen's listed Zapier, Make and n8n integrations cover the trigger and the call, and HeyGen MCP lets Claude or ChatGPT create videos without an API key. Code helps once you need custom retry logic at high volume.
Final Verdict
HeyGen is the pick because it is the cheapest presenter API here at $0.60 a minute on a digital twin, and the only one with listed Zapier, Make and n8n integrations plus an MCP route. Tavus wins replica realism. D-ID wins the flattest bill. Creatomate wins the no-code template layer around a presenter.
HeyGen's API starts at $5 with no subscription. Wire one CRM trigger to a single template and price ten videos before you scale.
Greetings! My name is Ayesha Shaheryar. My words have helped millions over the past two years. As a HeyGen expert and a writer, I am here to introduce tips and tricks to edit your next video in no time.
