Script · one voice per host · intro music · talking video
Write the episode, give each host a voice from 555 presets (or your own cloned voice), add a short intro theme and download clean clips for your editor. Want it on YouTube too? Turn the voices into a talking video of one host, or two people in the same frame.
Updated by the Kenerate AI team
On this page
Quick answer
An AI podcast generator turns a written script into a podcast episode spoken by AI voices, so you don't need a microphone or a recording booth. Kenerate AI does it in steps, not in one click: you write the script, voice each host in Voice Studio (from 15 credits per 1,000 characters), make an intro theme in Music Studio and join the clips in any audio editor. For a video podcast, Avatar Talk makes one host or two people talk to those voices, up to 10 minutes per video.
Checked 22 October 2026
HeyGen's AI podcast generator is the page that tops Google for this search. Here's how the two approaches differ.
| Kenerate | HeyGen AI Podcast Generator | |
|---|---|---|
| Document in, episode out | No — you write the script and voice each host | Yes — text, PDF, DOCX, PPT or a URL |
| Hosts | As many as you voice; two on screen with the Duo avatar engines | One or two AI voices |
| Preset voices | 555 (MiniMax, ElevenLabs, Qwen3, Kling) | 300+ |
| Languages | Up to 70 (ElevenLabs Eleven v3) | 177+ languages and dialects |
| Your own voice | Instant clone from a 3–15 s sample | Clone from a 30-second sample |
| Video version | Avatar Talk: one host up to 10 min, or two people from one photo | Lip-synced avatar host or a faceless video |
| Intro music | Made separately in Music Studio (12 credits a cue) | Background music, intros and outros built in |
| Price per episode | Shown before you generate: per character (voice), per second (video) | No prices on the generator page |
| Free start | 25 credits on sign-up, no watermark | Free plan; MP4 exports carry a watermark (its FAQ) |
HeyGen facts from its own AI podcast generator page, read 22 October 2026. Kenerate facts from the app's code.
Sources: Kenerate Voice Studio (engines, voices and limits from the app's code) · MiniMax Speech 2.8 HD on WaveSpeed · ElevenLabs Eleven v3 on WaveSpeed · ElevenLabs — text to speech documentation · Qwen3 TTS Voice Clone on WaveSpeed · ElevenLabs — Eleven v3 audio tags · Apple Podcasts — Content and subscription guidelines (§1.11 AI-generated content) · YouTube Help — Disclosing altered or synthetic content · Google Help — Generate Audio Overviews · HeyGen — AI podcast generator (read 22 October 2026)
Listen first
The sample on the left is an AI voice (nobody's real voice was copied). The clone on the right reads a line the sample never said. Clone only your own voice, or a co-host's with their clear consent.
Sample · 6 s
Upbeat Woman · English
Hi! I'm recording this in my kitchen, so excuse the echo. I just want to see if this actually sounds like me.
MiniMax Speech 2.8 HD · 0:06
Clone
Instant clone · English
Okay, that's honestly a little spooky. It even does the thing where I speed up when I get excited.
Qwen3 Voice Clone · 0:04
A 6-second “kitchen recording” becomes a reusable host voice. Qwen3 Voice Clone needs no training step and bills like normal speech: 15 credits per 1,000 characters.
More voices
Real Voice Studio outputs: a late-night host, a show opener, a chatty co-host, a designed radio voice and a station ID. Press play and follow the words.
Qwen3 TTS · Alibaba Qwen
You're listening to Night Shift, the show for people who do their best thinking after midnight. Grab something warm to drink. We're in no hurry.
Style: "calm late-night podcast host, relaxed, warm and unhurried"
ElevenLabs Eleven v3 · ElevenLabs
[excited]Welcome to the show! Today we're taking a plain block of text and turning it into a voice you'd swear was recorded in a studio. [whispers]And nobody has to know.
ElevenLabs Eleven v3 · ElevenLabs
[cheerfully]Hey! So I finally tried that new ramen place downtown, and honestly? Worth the forty-minute line.
Qwen3 Voice Design · Alibaba Qwen
Voice descriptionA warm, slightly husky woman in her forties with a gentle Southern American drawl, speaking slowly like a country radio host at dawn
Mornin', y'all. The coffee's on, the fields are quiet, and we've got three hours of the best old songs you ever heard.
MiniMax Speech 2.8 HD · MiniMax
You're listening to Riverside Radio. Ninety-eight point four. The best mix of yesterday and today.
ElevenLabs Multilingual v2 · ElevenLabs
It was a grey Tuesday in November when the letter arrived, and nothing in the village was ever quite the same again.
How to
Step 1
Mark every line with who says it (SAM:, ALEX:). Ten minutes is about 1,300–1,500 words. Write it yourself or draft it with Kenerate Agent, then fact-check it.
Step 2
In Voice Studio, choose a preset, design a voice from a description, or clone your own from a 3–15 s sample. Keep the same voice and settings for every episode.
Step 3
Generate one clip per turn (or per block of turns) in that host's voice. Each clip can hold up to 10,000 characters on MiniMax; replay and regenerate any line that sounds off.
Step 4
In Music Studio, ask Ken Music 1 for a ~30-second instrumental cue (12 credits), or use ElevenLabs Music v2.5 for an exact length.
See it in the app
Open Image Studio and type what you want to see — subject, style, lighting and mood.
Text to speech, emotions & tags, Hindi, voice design, cloning & chat
Chapters
Use cases
Where a scripted, AI-voiced show works well — and where you still want a human in the room.
One host asks, the other explains. Write both parts, give each its own voice, keep them the same every week.
A short weekly round-up read by one steady voice; regenerate a line when a fact changes.
Turn lessons into listenable episodes for learners on the move.
A narrator plus character voices, with Voice Design for anyone the presets don't cover.
Translate the script and voice it again: up to 70 languages on ElevenLabs Eleven v3.
Give your hosts a face with Avatar Talk — one presenter, or two people from one photo.
Engines
From the Voice Studio's code, 22 October 2026. Prices are in the facts above and the FAQ.
MiniMax
Best for
Most natural MiniMax voice; 7 emotions, (laughs)/(sighs) interjections, pronunciation dictionary
ElevenLabs
Best for
Most expressive; audio tags like [whispers], [laughs], [excited]
ElevenLabs
Best for
Lifelike and consistent over long reads
Alibaba Qwen
Best for
9 characterful voices plus a free-text style prompt
Alibaba Qwen
Best for
Describe a voice in words and hear it
Alibaba Qwen
Best for
Instant clone from a 3–15 s sample, no training
Controls
Contrast is what keeps a two-voice show easy to follow.
Pick host voices from different engines or genders so listeners never mix them up.
MiniMax voices. Keep the host near 1×; a slightly quicker co-host (1.05–1.1×) sounds livelier.
MiniMax: neutral, happy, sad, angry, fearful, disgusted, surprised — set it per line, not per episode.
MiniMax 2.8 reads 13 interjections like (laughs) and (breath); Eleven v3 reads tags like [laughs], [curious], [sarcastic].
Qwen3 TTS: keep a voice and steer it — “calm late-night podcast host, relaxed and unhurried”.
MiniMax offers WAV and FLAC at up to 44.1 kHz for editing; every clip downloads without a watermark.
Scripts
Open any of these in Voice Studio, then swap in your show's name and topic.
Cold open (host A)
Here's a number that surprised me this week: most people check their phone within ten minutes of waking up. So today, one question. Is that a habit, or is it a design choice?
Reply with a laugh (host B, Eleven v3)
[laughs] Okay, I'm checking mine right now, so I'm clearly not the right person to answer that. [curious] But go on. Who designed it?
Late-night host (Qwen3, style prompt)
It's just past midnight, the city's gone quiet, and we've got one story to get through. Pour something warm. We're in no hurry.
Add a style prompt such as “calm late-night podcast host, warm and unhurried”.
Sponsor-free ad break
Quick break. This show has no sponsors, just listeners. If it helped you this week, share it with one person who'd like it. Right, back to it.
Outro
That's the episode. Sources and notes are in the show description. Thanks for listening, and see you next week.
Tips
Most of the quality comes from the script and the edit, not the voice.
Write how people talk: short sentences, interruptions, “wait, really?” — not an article split in two.
Give the hosts different jobs: one asks the listener's questions, the other has the answers.
Two voices from different makers or genders are easier to tell apart in earbuds.
Generate one clip per turn so a fix means regenerating a line, not the whole episode.
Leave 0.3–0.5 s of silence between turns in the edit; AI clips start and stop cleanly, so add the breathing room yourself.
Check every fact before you publish — the voices will read mistakes just as confidently.
Clone only your own voice, or a co-host's with clear consent. Never a celebrity's or a stranger's.
FAQ
10 questions
No. Kenerate doesn't turn a PDF or a link into a finished two-host episode in one click. You write the script (or draft it with Kenerate Agent), voice each host in Voice Studio, make an intro theme in Music Studio and join the clips in an audio editor. It takes longer than a one-click tool, but you control every word, voice and pause.
Link to this answerYes. Give each host their own voice — for example a MiniMax voice for the host and an ElevenLabs voice for the co-host — and generate each host's lines in that voice. Then put the clips in order in your editor. For a video, the InfiniteTalk Duo engine in Avatar Talk animates two people in one photo, each speaking from their own audio track.
Link to this answerA 10-minute episode is roughly 1,300–1,500 words, about 8,000 characters. On MiniMax Speech 2.8 HD (30 credits per 1,000 characters) that's about 240 credits; on ElevenLabs Eleven v3 (60 per 1,000) about 480; on Qwen3 TTS (15 per 1,000) about 120. A Ken Music 1 intro cue adds 12 credits. Each clip rounds up to a whole credit. One-time packs start at $15 for 1,300 credits.
Link to this answerFree to start: a new account gets 25 credits, enough for about 800 characters on MiniMax Speech 2.8 HD — plenty to audition voices for your hosts. A full episode needs more credits. Voice Studio, Music Studio and Avatar Talk don't need a paid plan, just credits, and nothing carries a watermark.
Link to this answerYes. Upload or record a clean 3–15 second sample and Qwen3 Voice Clone reads your script in that voice straight away, for 15 credits per 1,000 characters. You can also save a clone for the MiniMax engines (480 credits once). Only clone your own voice or one you have clear permission to use.
Link to this answerYes, in Avatar Talk. For one host, add a portrait and the voice track: Kenerate Avatar runs up to 10 minutes at 7.5 credits a second (720p), SoulX FlashHead at 4.5 a second (480p). For two people, InfiniteTalk Duo takes a photo of both and one audio track per person, at 9 credits a second (480p). A 10-minute video costs thousands of credits, so try a 30-second clip first.
Link to this answerOn Apple Podcasts, yes: its content guidelines (section 1.11) say creators using synthetic voices or AI hosts must disclose it prominently in the content and in each episode's and show's metadata. YouTube asks for a label on realistic synthetic content, but not for cloning your own voice for a voice-over. A line in the intro and the show notes covers both.
Link to this answerEach Voice Studio clip holds up to 10,000 characters on MiniMax and ElevenLabs Multilingual v2, and 5,000 on Eleven v3 and Qwen3. Most two-host shows work best one turn per clip anyway. Avatar Talk videos run up to 10 minutes each; split longer episodes into parts.
Link to this answerPartly. In the chat it can draft the script, voice it with generate_speech and compose a theme with generate_music, and in Ask first mode it shows you the plan before it spends credits. It can join finished video clips with music (2–20 clips, up to 3 minutes), but it has no tool to stitch audio-only clips into one long episode — that last step happens in your editor.
Link to this answerGoogle's notebook tool makes a two-host audio discussion from documents you upload, and its own help page warns the result may contain inaccuracies. It's a summary of your sources, not a show you script. Kenerate is the other way round: you decide what each host says, which voice they have and how the episode sounds.
Link to this answerIn depth
What works, what to avoid, and how the pieces fit.
Two different tools hide behind the same search. One kind reads your document and writes and voices a conversation about it for you — a summary in podcast form. The other kind gives you studio-quality voices and leaves the show to you: you script it, cast the voices and edit it, without a microphone.
Kenerate is the second kind. That's slower than a one-click summary, but it's the version that sounds like a show with a format, recurring hosts and an intro, rather than two voices reacting to a PDF.
A script that reads well on paper often sounds stiff out loud. What helps:
Kenerate gives you clean, separate clips; the edit is yours:
For one host, Avatar Talk lip-syncs a portrait to the voice track: Kenerate Avatar for the cleanest 1080p presenter video, SoulX FlashHead for long episodes at the lowest rate. For two people, the Duo engines take one photo with both of them and a separate track per person. In “Together” mode you're billed for the longer track, so build both tracks to the same timeline in your editor — each with silence while the other person talks.
Video is where the credits go: speech for a 10-minute episode costs a few hundred credits, while a 10-minute Duo video at 480p costs 5,400. Render a 30-second test first.
The rules worth knowing before you publish:
Free to start. No watermark. Download every clip.
Tried it? Tell us how it went
