AI SDR tools are good at writing personalized emails, and that is now the problem: every inbox is full of them. Video from a real rep stands out, but no rep can record a hundred takes a day, and generic recorded videos lose the personal touch that made them worth sending. Avatar video closes that gap if the result looks like the rep and the pipeline is built for volume.
This guide covers how to add personalized video to an AI SDR product: what to personalize, how to structure the pipeline, what it costs per video, and the mistakes that make prospects stop watching.
What "personalized" should mean
The temptation is to generate a completely bespoke script for every prospect. In practice, a strong outbound video has a fixed spine and a few personalized slots:
- Opener: the prospect's first name and company, said in the first two seconds.
- Reason for reaching out: one specific line drawn from your research step (a hiring post, a funding announcement, a product launch).
- Value line: fixed per campaign, written and approved by a human.
- Call to action: fixed per campaign.
Keeping the spine fixed has three benefits. Messaging stays on brand, review is easy because only the slots change, and you can measure which template works instead of comparing thousands of one-off scripts. Your LLM fills the slots; it does not rewrite the pitch.
The pipeline
1. One clip per rep
Each rep records a single clip of 10 to 120 seconds, ideally looking at the camera with natural expression and steady light. Your app uploads it and creates an avatar. If the clip contains at least 10 seconds of speech, the voice is cloned from the clip itself; a separate 1 to 2 minute voice sample gives the best clone.
curl -X POST https://pitchtwin.ai/api/v1/avatars \
-H "Authorization: Bearer pt_live_your_key" \
-H "Content-Type: application/json" \
-d '{ "name": "Maya - AE", "video_url": "upload:file_abc123" }'
The response is 202 with status processing. Poll GET /avatars/:id until it is ready, then store the avatar ID on the rep's profile. Because each avatar is pinned to its engine version, the rep's face will not drift between this quarter's campaign and the next.
Since the avatar is built from real video of the rep and keeps that footage, the prospect sees the rep's actual skin, hair, lighting and expressions. That matters in outbound, where any hint of a synthetic face reads as spam. See why AI avatars look plastic for the detail.
2. Per-prospect scripts with pauses
Use script mode so you can place short pauses after the personalized slots. A pause after the name gives the opener the rhythm of a real greeting.
const API = "https://pitchtwin.ai/api/v1";
async function createProspectVideo(avatarId: string, p: {
firstName: string;
company: string;
reason: string;
prospectId: string;
}) {
const res = await fetch(`${API}/generations`, {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.PITCHTWIN_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
avatar_id: avatarId,
script: [
{ text: `Hi ${p.firstName},`, pause_after_ms: 400 },
{ text: `I was looking at ${p.company} this morning. ${p.reason}`, pause_after_ms: 300 },
{ text: "We help teams like yours cut onboarding time for new reps.", pause_after_ms: 300 },
{ text: "Worth a quick call next week? Link is right below.", pause_after_ms: 0 },
],
voice: { speed: 1.0, language: "en" },
webhook_url: `https://app.example.com/hooks/pitchtwin?prospect=${p.prospectId}`,
}),
});
const gen = await res.json(); // { id, status: "queued" }
return gen.id as string;
}
Store the returned generation ID against the prospect before doing anything else, so a retry in your own job runner does not create a duplicate video. Writing tips for these slots are in writing scripts for AI avatars.
3. Webhook, then copy to your storage
When the video is done, PitchTwin POSTs generation.succeeded (or generation.failed) to your webhook_url, signed with the PitchTwin-Signature header. Verify the signature, then fetch GET /generations/:id for the video_url. Webhooks are delivered once, so keep a polling fallback for anything still pending after a while. Details are in the webhook docs.
Video and audio links are signed and expire after 24 hours. Outbound emails get opened days later, so copy the MP4 to your own storage or CDN immediately and serve it from there.
4. Send a link, not an attachment
Do not attach the MP4 to the email. Generate a thumbnail from an early frame (ffmpeg works fine), put the prospect's name on it, and link to a lightweight landing page with the player and the calendar link. You control the page, so you can track plays and watch time in your own analytics.
The response also includes words with word-level timings. Use them to render captions on the landing page, since many people watch with the sound off.
What it costs per video
PitchTwin uses prepaid tokens that never expire, with no seats or subscription. One token is one second of video with your own audio; 1.5 tokens per second when PitchTwin voices the script. For a 30-second prospect video:
| Pack | Price | 30 s, PitchTwin voice (45 tokens) | 30 s, your audio (30 tokens) |
|---|---|---|---|
| Starter | $9 / 600 tokens | about $0.68 | $0.45 |
| Growth | $39 / 3,000 tokens | about $0.59 | $0.39 |
| Scale | $149 / 12,000 tokens | about $0.56 | about $0.37 |
Failed generations are refunded in full automatically, so you only pay for videos you can send. If your tool already generates speech through ElevenLabs or another provider, audio mode is the cheaper rate. Full numbers are on the pricing page.
Shorter videos are both cheaper and usually better. Twenty to forty seconds is enough for a name, a reason and an ask.
Mistakes that kill reply rates
- Personalizing too late. If the name comes after a generic intro, the video feels like a template. Put it first.
- Mispronounced names. Spell difficult names phonetically in the
textfield and keep a per-prospect override your users can edit. - Too many slots. Every extra variable is another place the LLM can produce something odd. Two slots is plenty.
- No human review. Let users approve a batch before it sends. A preview of the script is enough; they do not need to watch every video.
- Using a rep who has left. Tie each avatar to an active user and delete it when they leave. Consent matters here too: the rep has to agree to their avatar being used, and that agreement should be recorded. More in consent and likeness rights for AI avatars.
White-label delivery
Your customers are selling their product, not yours, and certainly not your video vendor's. PitchTwin output has no watermark and no PitchTwin branding, and videos go to your webhook, so the whole experience stays inside your product. The personalized video API page covers the model in more detail.
Try it
Record a 30-second clip of yourself, create an avatar in the dashboard and use the Playground to generate a test outbound video without writing code; it shows the equivalent cURL and Node request. Sign up for 60 free seconds, no card required.