The idea behind Seagull
Local vs cloud voice cloning on a Mac
Almost every voice cloning tool is a website that uploads your voice to someone else's GPU and bills you monthly. Seagull Studio takes the other path: a native macOS app that clones a voice from a 10-second sample and generates speech on your own Apple Silicon Mac, offline after the model is downloaded, for $11.99 once. This page explains what that trade actually buys you, and where the cloud still wins.
What does "local voice cloning" actually mean?
Voice cloning has two parts: analysing a reference recording to capture how a voice sounds, and synthesising new speech from text in that voice. In a cloud service both steps happen on a server. You upload a sample, the service stores a voice profile, and every sentence you generate is a request that leaves your machine and comes back as audio.
Local voice cloning does both steps on your own hardware. Seagull Studio downloads a model once, then runs it on your Mac's Apple Silicon chip. Your voice sample and your script are inputs to a process on your own computer, the same way a photo is an input to an image editor. There is no upload step and no per-sentence request.
The practical test: pull the Wi-Fi plug and try to generate a sentence. A cloud tool stops working. A local tool keeps going.
Why did I build Seagull Studio this way?
One trigger was subscriptions spreading to everything. I understand why cloud voice tools have to work that way. Renting GPUs to run large text and audio models is expensive, and a one-time purchase on top of that would be a losing business. If your product is someone else's hardware, you have to meter it.
The other trigger is the machine sitting on the desk. A normal home computer is now powerful enough to do this work. People already paid for that hardware; they should be able to use it for something demanding, instead of watching it idle while a monthly plan rents the same capability from a datacenter.
The industry keeps pushing everyone toward a handful of centralized services. I do not like that kind of dependence. A voice tool should be something you own, not a relationship you have to keep paying to keep working. Anyone who wants to stay independent of the AI platforms should have a way to do that.
A voice is also not the same kind of data as a document. It identifies you, it can be used to impersonate you, and once it is stored on someone else's infrastructure you are trusting a retention policy you cannot inspect. Seagull Studio exists so that using your own voice does not require that trust: the app carries Apple's Data Not Collected privacy label on its Mac App Store listing, and synthesis happens on the machine in front of you.
The pricing follows from the same idea. If the compute is yours, there is nothing to meter. That is why Seagull Studio is a one-time $11.99 unlock with 20 free generations to try first, instead of a monthly plan with a credit balance.
What is Seagull Studio becoming?
I want Seagull Studio to be the app that keeps picking up the modern models that can actually generate audio on a laptop. The six models in the app now are a starting set, not a closed catalogue. As new local models appear, I will test them, keep the ones that earn their disk space, and look for combinations that work better together than any single model on its own.
I also want that work to stay in conversation with the people using the app. If something is missing — a language, a workflow, a model you already run elsewhere — write to seagull.meow.sup@gmail.com or @seagull_meow. I am ready to move the product in the directions that turn out to matter.
What stays on your Mac, and what still needs the internet?
Being honest about this matters more than the slogan. Seagull Studio is not an air-gapped appliance; it is a normal Mac app that needs the network for two specific things.
| Step | Runs on your Mac | Needs the internet |
|---|---|---|
| Recording or importing a voice sample | Yes | No |
| Voice cloning and speech generation | Yes | No |
| Generation history and exports | Yes | No |
| Downloading a model the first time | Stored locally after download | Yes |
| Purchase, restore, Family Sharing | — | Yes, through the App Store |
Where are cloud services genuinely better?
Anyone who tells you local is better in every dimension is selling something. Cloud voice platforms are ahead in several real ways, and if one of these is your requirement, they are the right tool:
- Top-end voice quality. Server GPUs run models far larger than anything that fits comfortably on a laptop, and the best cloud presets still set the bar for polished English narration.
- Any device, any OS. A browser tab works on Windows, on an iPad, on a borrowed laptop. Seagull Studio needs your Mac.
- APIs and automation at scale. If you are generating audio from a backend or a product, you want an HTTP API with concurrency guarantees. Seagull Studio exposes local MCP tools for agents on the same machine, which is a different job.
- Teams and shared voice libraries. Seats, permissions and shared voice slots are subscription features, and Seagull Studio has no equivalent.
- Breadth of languages and extras. Large platforms cover more languages than local models and add dubbing studios, transcription and voice design around the core feature.
What does running models locally cost you?
The trade is not free. Three costs are worth knowing before you buy:
Hardware. Seagull Studio requires macOS 14 or later and an Apple Silicon Mac. Intel Macs are not supported. Bigger models want more unified memory, and generation speed depends on your chip rather than on someone else's queue.
Disk space. The app itself is 377.5 MB, but the models are the real weight: 6 selectable models from about 0.7 GB to 13 GB. You choose what to download.
Tuning. With a cloud service you pick a preset voice and it sounds good immediately. Locally you pick a model, and models differ: one is fast and English-only, another covers 23 languages but downloads 13 GB, the compact multilingual option reaches up to 31 languages in about 0.7 GB. Getting the best result means trying a couple of them on your own script.
Generation examples are coming soon — the same script through a few on-device models, so you can hear the trade-offs instead of reading about them.
How does the money work out?
Cloud voice services meter output. You buy a monthly credit balance, credits convert to minutes of audio, and unused balance either partially rolls over or expires. Seagull Studio charges once and does not meter anything: after the unlock, the only limit is how long you are willing to let your Mac work.
| Option | Entry price | What you get |
|---|---|---|
| Seagull Studio | $11.99 once | Unmetered local generation, 20 free generations before you buy |
| ElevenLabs | $6 / month (Starter) | 30,000 credits, roughly 30 minutes of audio, instant cloning and commercial rights; the free tier excludes commercial use |
| Fish Audio | $15 / month (Plus) | 250,000 credits, up to about 200 minutes of audio; the free tier is personal, non-commercial use only |
In plain arithmetic, Seagull Studio costs less than two months of the cheapest ElevenLabs plan and less than one month of Fish Audio Plus. That comparison is only fair for someone doing steady personal or small-project work on a Mac. It is not fair if you need an API, a team, or the absolute best-sounding English voice — then you are buying different things.
Who should choose which?
Choose local (Seagull Studio) if your voice or your client's voice should not be uploaded anywhere; you work on planes, trains or bad hotel Wi-Fi; you narrate regularly and dislike watching a credit balance; or you simply want to own the tool instead of renting it.
Choose a cloud service if you need the highest-quality English preset voices, an API for a product, collaboration with a team, access from non-Mac devices, or a language local models do not cover yet.
There is also a reasonable middle path: keep a cloud subscription for the occasional job that needs it, and do everyday work locally. A $11.99 one-time purchase does not have to replace anything to pay for itself.
How does it work in practice?
Three steps, all on your machine. Record at least 10 seconds of speech in Seagull Studio, or import an existing file as WAV, MP3, M4A, AAC, AIFF or CAF. Pick a model and write your script — the app can detect the language itself, and it preprocesses Russian stress marks and Japanese kana. Generate, listen, keep the result in history, and export as WAV or MP3.
Every one of those steps happens between your microphone, your disk and your chip. That is the whole idea.
Questions, answered.
Is local voice cloning as good as cloud voice cloning?
Not always. Cloud services run much larger models on server GPUs and usually win on the most polished English narration. Local models have closed a lot of that gap, and for many voices the difference is small, but if you need the single best-sounding result you should compare both on your own script before deciding.
Does Seagull Studio need an internet connection?
Only for downloading models and for App Store purchase and restore. Once a model is on disk, generation runs on your Mac and works offline.
How long a voice sample do I need?
At least 10 seconds. A clean, quiet recording of natural speech works better than a long noisy one.
Why does a local app need Apple Silicon?
The models run on the Mac's unified memory and neural hardware. Seagull Studio requires macOS 14 or later on an Apple Silicon Mac; Intel Macs are not supported.
How much disk space do the models take?
Seagull Studio ships with 6 selectable models ranging from about 0.7 GB to 13 GB. You download only the ones you want to use, and you can remove them later.
Is a one-time purchase really cheaper than a subscription?
For steady personal use, yes. Seagull Studio costs $11.99 once with no recurring fee, while cloud voice services bill monthly and meter output in credits or minutes. Cloud plans remain a better fit if you need an API, team seats, or browser access from any device.