Record your own face and voice one time, then write a fresh script for each prospect and let the models do the talking. Clone your voice, animate your photo or your clip with lipsync, and review every video before it goes out, because the sending and the responsibility stay with you.
A cold email with a short video of a real person talking in it gets read differently than a wall of text, but recording one video per prospect is a chore nobody keeps up for long. The AI version splits the work in two: you record your face and voice once, then write a new script for each person and let models do the talking. This page covers how that works on Picasso IA, which models fit, and where it stops being useful.
A text template is easy to spot. By the third line the reader knows a sequence sent it, and the name in the greeting does not change that. A short video of a person speaking to them, mentioning their company and one specific thing you noticed, is harder to mistake for a mass send, mostly because it still takes effort to look at.
The catch has always been the effort on your side. Filming, retaking and exporting twenty videos in an afternoon is a day of work, and by video twelve you sound tired. Generating the talking part from your own face and voice moves the effort to the script, which is where the personalization lives anyway. Picasso IA has 502 models in the catalog, including twelve in lipsync and twenty-four in text-to-speech, so you can build the presenter from parts instead of depending on one closed tool.
One boundary before anything else: this page is about your own face and your own voice. Not a colleague's, not a founder's, not anyone you do not have permission from. The whole idea only works honestly if the person on screen is the person sending.
There is no single right setup. What changes between them is what you have to record, how much of the footage is really you, and how much control you keep over the words. This table compares the options that exist on the platform today.
| Setup | What you provide | Strongest at | Watch out for |
|---|---|---|---|
| Your photo plus a cloned voice, animated with Kling Avatar v2 | One clear photo and a short audio file | The fastest route from script to talking video | Audio file limited to 5 MB, so keep scripts short |
| Your own recorded clip, re-voiced with Lipsync 2 | A short MP4 of you and a WAV of the new line | Real head and body motion that is actually yours | Needs a decent source clip, and mouth sync is only as good as that footage |
| A scripted presenter in Avatar IV | A script, an avatar ID and a voice ID | Typing a message and getting a presenter without a recording | Avatar and voice IDs come from HeyGen's own lists, so check what you can use |
| Cloned voice over your screen or slides | Your voice sample and a screen recording | Demos where your face is not the point | No face on screen, so less of the personal touch |
The first two are the ones that deliver the title promise most directly, because the face is yours in both. The third is useful when you want to write and ship without recording anything, but read its model page before assuming an avatar made from you is available to it. For a deeper look at the talking-photo route, see the AI talking photos page.
This is the only recording session, so it is worth doing properly. Sit near a window, face the light, and put the camera at eye level. Read a short passage in the tone you would use on a real call: friendly, a little slower than you think is necessary, mouth clearly open on the vowels. Record a clean take of about a minute, then record a second one standing or sitting a bit differently, so you have a spare.
From that session you pull three things. A still frame of your face for photo animation. The video clip itself, for lipsync. And the audio, for voice cloning. The voice cloning model on Picasso IA, minimax voice-cloning, accepts MP3, M4A or WAV from 10 seconds up to 5 minutes, with an option to reduce background noise, and then lets you reuse the cloned voice across speech runs. Chatterbox from resemble-ai is a different route: it clones from a few seconds of reference audio inside the same run and has an emotion setting.
Quiet matters more than quality. A cheap microphone in a silent room beats an expensive one next to a fan, because the cloning model copies whatever is in the sample, hum included. Do the recording before you open any model.
Listen to the first cloned line out loud before you build anything on it. If it sounds like you on a bad day, redo the sample. If it sounds like a stranger, redo the sample with less background noise and a steadier pace.
This is the loop you will repeat for every prospect once the recording exists. It runs entirely in the toolkit, where image, video, voice and lipsync models sit in one place.
Every prospect is one pass through these steps. There is no queue that takes a spreadsheet and returns a folder of videos, and nothing on this page should be read as saying there is.
The point of personalizing is that the prospect can tell you looked. So spend your edits there, and keep everything else fixed so you are not reviewing a different video every time.
Change the first sentence and the specific observation: a product they launched, a role they just started, something in a post of theirs that you have actually read. Keep the offer sentence and the closing question the same, or very close. That gives you a stable core that you have already checked, and a small, clearly personal part that you check each time.
Keep the video short enough that nobody scrubs ahead. Write for the ear, with short sentences and plain words, because a tongue-twister of a company name will trip the voice model just as it trips a person. When a name is unusual, spell it the way it sounds in the script and listen again. If you are tempted to add a long product pitch, the better move is a link in the email and a video that only opens the door. For longer demos, the AI product demo videos page is the place to start.
You can also reuse the same voice to record the follow-up. A second short video a week later, in the same face and voice, lands better than a text nudge, as long as it says something new.
Plenty of outreach does not need this, and some of it is made worse by it. These are the limits that matter in practice:
For quick reach to existing contacts, a plain email you wrote yourself is sometimes the better message. Video earns its place when it adds something text cannot: your voice, your expression, a thirty-second sense of who is writing.
Yes, that is the intended use. You record yourself once, clone your voice from that sample, and then generate talking videos from your photo or your clip with a new script each time. The models run in the toolkit on Picasso IA. The limit is on the likeness: it should be you, or a fictional character you created, never another real person without their permission.
For the voice, minimax voice-cloning and Chatterbox from resemble-ai both clone a voice from a sample. For the face, Kling Avatar v2 animates a photo with your audio, and Lipsync 2 re-times your own recorded clip to new speech. Avatar IV from heygen produces a scripted presenter from an avatar ID and a voice ID. Model availability changes, so confirm each one in the catalog before you plan around it.
Not today. The toolkit runs one generation at a time, and there is no feature that takes a spreadsheet of prospects and renders a video for each. You can reuse your cloned voice and your photo as many times as you like, but each video is its own run that you write, generate and review. If your plan depends on hundreds of fully automatic sends, this is not the right tool for that part of it.
It is the safer choice, and in some places it may be a requirement, so check the rules that apply to you. A short line in the email, such as "I made this video with AI from a recording of me", costs you little. The alternative, letting someone believe you filmed it live for them, tends to backfire the moment they notice the loop in your expression or a slightly off pause.
You do. Rules about unsolicited messages, consent, opt-outs and data use differ by country and by channel, and the sender is responsible for following them, not the model that made the video. This page is general information and not legal advice. If you send at any volume, have a qualified professional look at your process, and check each platform's current requirements before you post or message.
The minimax voice-cloning model accepts a recording from 10 seconds up to 5 minutes in MP3, M4A or WAV, under 20 MB, with an optional noise reduction setting. Chatterbox works from a few seconds of reference audio. More is not automatically better: a clean minute in a quiet room usually beats five minutes with a fan running behind you.
Photo animation, with Kling Avatar v2, needs only one still image and an audio file, so it is quick, but the motion is generated. Lipsync 2 starts from your real video and adjusts the mouth to the new audio, so the head and body movement is actually yours. The second route takes a bit more recording up front and usually looks more personal.
No. Use your own face and voice, or fictional characters. Cloning a real person without permission is a trust problem first and often a legal one second, and it is also the quickest way to get a prospect to report you. If a team wants this, every person who appears should record their own sample and agree to it.
Some of the speech models describe multilingual output, and the model pages mention writing the script in another language with a cloned voice. Quality varies by language and by voice, so test with a short line first and ask a native speaker to listen before you send anything to a real prospect, especially for names and technical terms.
Generations are paid in credits, and the cost depends on the model and the settings you choose, so any figure written here would go stale. Video costs more than audio or images. New accounts get free credits to try the workflow on a few prospects, and the pricing page is where the current plans are listed.
Ready to try it on one real prospect first? Open the Picasso IA toolkit, record your sample, and build a single video before you decide how far to take it.
Choose your language
