Business

White-label AI avatars: what B2B tools need

White-label AI avatars for B2B tools: branding, delivery, storage, tenant isolation, consent and disclosure, with a checklist and a Node webhook example.

PitchTwin team, , 6 min read

If you sell a B2B product that generates avatar videos, your customers expect the videos to look like they came from you. A vendor watermark, a vendor-branded player, or a download link on someone else's domain undermines that. But "white-label" means more than removing a logo: it covers how videos are delivered, where they live, how tenants are kept apart, and how consent is handled.

This post lists what a white-label avatar setup needs in practice, and how to build each part on top of an avatar API.

What white-label actually means

For an end customer, a white-labeled video should be indistinguishable from something your product made itself. That breaks down into five requirements:

  1. No third-party branding in the video, on the file, or in any player.
  2. Delivery into your systems, not a vendor's dashboard or email.
  3. Hosting under your control, on your storage and your domain.
  4. Tenant isolation, so one customer's avatars and costs never mix with another's.
  5. Consistency over time, so an avatar looks the same next month as it does today.

There is one thing white-label does not mean: hiding that a video is AI-generated from the people who watch it. More on that below.

1. No branding in the output

The baseline is an output file with no watermark, no intro or outro card, and no vendor metadata your customer would see. PitchTwin is white-label by default: there is no watermark and no PitchTwin branding, and the output is a plain MP4.

When you evaluate any provider, check the actual file rather than the marketing page. Render a test video, download it, and look at the first and last frames and the file metadata.

2. Delivery into your systems

Video generation is asynchronous. Your app sends a request, the provider renders, and you need to know when the video is done. A white-label setup should deliver results to your endpoint, so you control what the customer sees and when.

With PitchTwin, you pass a webhook_url when you create a generation:

curl -X POST https://pitchtwin.ai/api/v1/generations \
  -H "Authorization: Bearer pt_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "avatar_id": "av_3f9c1e7a2b4d6e80",
    "text": "Hi Dana, here is a quick walkthrough of your new workspace.",
    "webhook_url": "https://app.example.com/webhooks/pitchtwin"
  }'

The response is 202 with { id, status: "queued" }. When rendering finishes, your endpoint receives a POST with type set to generation.succeeded or generation.failed.

Webhooks are delivered once, so keep a fallback: poll GET /generations/:id for any job that has not reported back after a reasonable wait. The webhooks section of the docs covers the payload.

3. Hosting under your control

The video_url returned by PitchTwin is a signed link that expires after 24 hours. That is deliberate: it is a handoff, not a hosting service. Copy the MP4 into your own storage as soon as the webhook arrives, then serve it from your domain or CDN.

Here is a Node handler that verifies the webhook signature, then copies the video into S3:

import crypto from "node:crypto";
import express from "express";
import { S3Client, PutObjectCommand } from "@aws-sdk/client-s3";

const app = express();
const s3 = new S3Client({});
const SECRET = process.env.PITCHTWIN_WEBHOOK_SECRET!;

function verifySignature(rawBody: Buffer, header?: string): boolean {
  if (!header) return false;
  const parts = Object.fromEntries(header.split(",").map((p) => p.split("=") as [string, string]));
  if (!parts.t || !parts.v1) return false;
  const expected = crypto
    .createHmac("sha256", SECRET)
    .update(`${parts.t}.${rawBody.toString("utf8")}`)
    .digest("hex");
  const a = Buffer.from(expected);
  const b = Buffer.from(parts.v1);
  return a.length === b.length && crypto.timingSafeEqual(a, b);
}

app.post("/webhooks/pitchtwin", express.raw({ type: "application/json" }), async (req, res) => {
  if (!verifySignature(req.body, req.header("PitchTwin-Signature"))) {
    return res.sendStatus(400);
  }

  const event = JSON.parse(req.body.toString("utf8"));
  if (event.type === "generation.succeeded") {
    const video = await fetch(event.data.video_url);
    await s3.send(
      new PutObjectCommand({
        Bucket: "customer-videos",
        Key: `videos/${event.data.id}.mp4`,
        Body: Buffer.from(await video.arrayBuffer()),
        ContentType: "video/mp4",
      })
    );
  }
  res.sendStatus(200);
});

The signature is a hex HMAC-SHA256 of ${t}.${raw_body}, using the webhook secret from Dashboard → API keys. Verify it against the raw request body, before any JSON parsing changes the bytes. A full walkthrough in Node and Python is in how to verify HMAC webhook signatures.

Copy the audio_url and words timings too if you build captions or transcripts. For production, you may also want to check the t timestamp against your clock and reject old events.

4. Tenant isolation

Your customers are separate businesses. Their avatars, voices, and usage should be kept apart in your system even if they share one provider account.

Map avatars to tenants

Store the provider's avatar ID against your own tenant ID. Use the name field on POST /avatars for something you can trace back, such as your tenant and user IDs, so logs make sense later.

Delete cleanly

When a customer leaves, or a person withdraws from an avatar, call DELETE /avatars/:id. On PitchTwin this also deletes the cloned voice. Build this into your offboarding flow rather than leaving it to a manual cleanup.

Track cost per tenant

Each finished generation returns tokens_charged. Record it against the tenant that triggered it. That gives you per-tenant cost for billing, margin reporting, and abuse detection. PitchTwin has no seats or subscriptions, so one prepaid balance can serve every tenant while your own database does the accounting.

5. Consistency over time

If an avatar subtly changes after a provider ships a new model, your customer sees their sales rep or trainer look different, and they will ask you why. On PitchTwin, each avatar is pinned to its engine version, so there are no silent upgrades and faces do not drift.

Quality matters here as well. Because PitchTwin builds avatars from real video of the person and keeps that original footage, results look 1:1 like the person, which is what a customer expects when the avatar carries their own face.

White-label shifts responsibility toward you, because your customers see your product, not your vendor's.

Consent. PitchTwin's Terms require that every person in an avatar has agreed to it. Build an explicit consent step into your avatar creation flow and keep a record of it. For more, see consent and likeness rights for AI avatars.

Disclosure to viewers. Removing vendor branding is not the same as hiding that content is synthetic. In the EU, Article 50 of the AI Act (Regulation (EU) 2024/1689) requires deployers who publish AI-generated or manipulated video that resembles real people to disclose that it is artificially generated, and it sets machine-readable marking duties for providers of generative systems. The deepfake disclosure obligation applies from 2 August 2026, and a summary from the Institute of Directors in Ireland notes a transitional period for some provider marking duties. This is not legal advice; check with counsel for your markets.

In practice, give your customers a simple way to add a disclosure line to the email, landing page, or player where the video appears.

Checklist

RequirementWhat to checkHow to handle it
No brandingWatermark, intro cards, metadataInspect a real output file
DeliveryResults reach your endpointwebhook_url plus polling fallback
HostingLinks expireCopy MP4 to your storage on arrival
SecurityWebhooks are authenticVerify PitchTwin-Signature on the raw body
Tenant isolationAvatars and costs per customerMap IDs; record tokens_charged
OffboardingAvatar and voice removalDELETE /avatars/:id
ConsistencyFaces do not changeEngine version pinned per avatar
ConsentEvery person agreedExplicit step in your flow, stored
DisclosureViewers know it is AICustomer-facing disclosure option

Try it

Sign up for 60 free seconds of video with no card, render a test video, and inspect the file yourself. When you are ready to wire up webhooks and storage, start with the quickstart and check the pricing page for pack sizes.

Keep reading