Demodokos Foundry 2.0: Speech v4, the Biggest Speech Update Since Launch

Product Update AI Voice Local AI Hardware TTS

Demodokos Foundry 2.0 introduced the Speech v4 model generation, the largest advance to Foundry's speech system since launch. The release rebuilt the speech engine, added support for more graphics cards (down to 4GB of VRAM) and refreshed the whole interface. Setup now shows you what it is downloading instead of stalling without a word.

Key takeaways

  • The new Speech v4 model generation keeps a voice consistent all the way through a clip, across every style and intensity, so it stops drifting or breaking character mid-line.
  • Foundry runs on more hardware now: AMD and Vulkan support got a real upgrade, and a new Medium model runs on as little as 4GB of VRAM.
  • The mixer and track visuals were rebuilt from the ground up, so you can finally see what your audio is doing, at much higher performance.
  • Setup is more reliable: the Package Manager now shows downloads, retries, and model inventory clearly instead of stalling silently.
  • Generation still runs locally on your own GPU, unlimited and without credits, and the Creator plan is still $12.00/month when billed annually.

What changed in Demodokos Foundry 2.0?

Foundry has always been a local AI audio suite for music, speech, voice cloning, and timeline editing that runs on your own Windows PC. Foundry 2.0 keeps all of that and improves almost every part of it. Speech v4 is the new model generation inside this application release. The table lists each change and what it means for your work.

Area What changed What it means for you
Speech engineNew Speech v4 model familyVoices hold character across styles and intensities
HardwareMedium model for low VRAMRuns on GPUs with as little as 4GB
HardwareBetter AMD / Vulkan handlingCleaner detection and setup when CUDA is not available
InterfaceFull UI and branding refreshCleaner windows and clearer visual feedback
MixerRebuilt track visualsSee your audio in detail, at higher performance
SetupImproved Package ManagerDownloads, retries, and model inventory are visible
MusicCreative AI on-demand loadingFaster startup, richer generation controls

What is the Speech v4 model generation?

Speech v4 is Foundry's voice model generation introduced in application version 2.0. Its main improvement is speaker consistency. It holds the same voice and the same character throughout a clip, even when the delivery shifts from calm to angry to a whisper. Earlier models could drift partway through a long line, which meant regenerating and hoping the next take held together. Speech v4 is built specifically to stop that.

Long work gains the most. If you narrate audiobooks, voice a cast of game characters, or run a faceless channel, a voice that stays in character across an entire paragraph saves you the tedious loop of regenerate, listen, regenerate again. The new native speech and TTS stack also improved batching, queueing, and the speech document workflow, so long jobs move more smoothly and stall less often.

The update also ships a refreshed demo voice library built to the new Speech v4 quality bar, plus a new v4 zvoice format for cloned and custom voices. If you have built voices in an older format, the new demo voices let you hear what the current system can do.

Can you run Demodokos Foundry on 4GB VRAM?

Yes. Foundry 2.0 adds a Speech v4 Medium model that runs on hardware with as little as 4GB of VRAM. Before this, the practical floor was higher, which shut out a lot of laptops and older cards. The Medium model brings the speech system to machines that could not load the larger models at all.

The Medium model is a full working option for people without a high-end GPU, who until now were told, essentially, to come back after an upgrade. If your card sat just under the old requirement, V4 is worth a fresh look. You give up some quality and speed; in return, it runs on your card at all.

Does Demodokos Foundry work on AMD GPUs?

AMD and Vulkan support got a real upgrade in Foundry 2.0. The update improves AMD and Vulkan detection, VRAM reporting, backend verification, and how setup behaves when CUDA is not available. In plain terms, if you are on an AMD card, Foundry does a much better job of recognizing your hardware and configuring itself correctly instead of assuming an NVIDIA CUDA path.

Foundry is still tuned first for Windows with an NVIDIA GPU, and that remains the smoothest experience. But the Vulkan work in Foundry 2.0 widens the door for people who were previously stuck at the setup screen. If a Vulkan backend failed to verify for you in the past, this is the release to try again on.

What changed in the mixer and interface?

Foundry 2.0 delivers a full interface refresh and a rebuilt mixer with new track visuals. The visuals are graded from Simplified to Accurate, so you can pick a lightweight view or a detailed one, and both run much faster than before. You can now watch what a track is doing in real time while you listen to it.

The wider interface refresh cleaned up window chrome, visual feedback, and the overall desktop experience. What Foundry produces stays the same. Working in it for hours at a stretch is more pleasant.

Is setup more reliable now?

Setup and startup are noticeably more reliable in Foundry 2.0. The improved Package Manager now surfaces downloads, retries, model inventory, and runtime setup clearly, instead of running hidden blocking downloads in the background. When something needs to download or a step fails, you see it and you can act on it.

Startup readiness also improved, so the app is less likely to report itself ready before its core services actually are. Alongside that, the release fixes a wide range of edge-case bugs across installation, model loading, document editing, style generation, and GPU or backend selection. You will less often sit wondering whether the app is working or stuck.

Music and Creative AI improvements

Foundry 2.0 continues to expand Foundry's music and creative tooling. Creative AI now loads on demand, which trims startup time and frees resources until you actually need the assistant. Caption and lyric workflows improved, generation controls got richer, and project and preset handling is more reliable, including reset, repair, append-track and the clip library.

If you build full tracks in Foundry, you will notice it in project handling and in the extra generator controls. For a closer look at how music generation fits into the wider suite, see how local AI music generation compares to Suno and Udio.

Why does local AI audio still matter here?

Every improvement in Foundry 2.0 and the Speech v4 generation lands on top of the same principle Foundry started with: it all runs on your machine. Your voice samples, scripts and generated audio are not uploaded to anyone's server. Generation is unlimited at a flat price, so there is no queue to wait in and nothing is charged per song.

A cloud service needs your files on its servers, however good its models get. Speech v4 brings stronger voices and Foundry 2.0 runs on more hardware, while your files stay on your PC. For the full picture of what the suite does end to end, see everything Demodokos Foundry can do.

Frequently asked questions

Do I have to pay extra for Speech v4?

No. Speech v4 is included in Foundry 2.0 and your existing subscription. The Creator plan is still $12.00/month when billed annually, and there is a 7-day free trial through PayPal with no charge on day one.

Will my existing voices and projects still work?

Yes. Foundry 2.0 adds a new v4 zvoice format and a refreshed demo voice library, but your existing projects continue to work, and so do the voices you already built.

What are the minimum requirements after the update?

Foundry runs on Windows 10 or 11 with an NVIDIA GPU as the primary path. The new Medium model runs on as little as 4GB of VRAM, and AMD and Vulkan support improved for cards where CUDA is not available.

Does the new speech model fix voices drifting mid-clip?

That is exactly what it targets. The Speech v4 model family is built to keep a voice consistent across styles and intensities, so it holds character through longer lines instead of drifting partway through.

How do I get the update?

Download the stable setup from demodokos.com and it pulls the latest version straight from Demodokos. If you already have Foundry installed, it updates to the current build for you.

Try Speech v4 in Foundry

If you are already a Foundry user, update and start with Speech v4 on a long line you know used to drift. That is where the difference shows up fastest. If you have not tried Foundry yet, or your GPU used to be too small to run it, Foundry 2.0 is a fair reason to take another look. Hear what local AI audio sounds like on your own machine. No charge today, cancel anytime during the trial.

Try Foundry Free for 7 Days

No charge during the trial. Cancel anytime.

More from the blog

OpenAI Is Retiring Its Text to Speech Models on January 6, 2027. What to Use Instead

OpenAI turns off tts-1, tts-1-hd and gpt-4o-mini-tts on January 6, 2027. What the Realtime replacement changes, what speech costs now and where to move. OpenAITTS 3 min

Qwen3-TTS, Chatterbox or Kokoro? Free Local Text to Speech Compared (2026)

Qwen3-TTS, Chatterbox and Kokoro run on your own PC at no cost (Chatterbox was retired in August 2026). We compared them on their own project pages and priced two real jobs against Demodokos Foundry. Local AITTS 7 min

AI Voices You Can Download and Keep on Your Own PC

Downloadable AI voices stay on your own PC, where nobody can withdraw them. How they work, how they compare with cloud libraries, and what they cost. Voice LibraryAI Voice 7 min

AI Dubbing Software That Runs on Your Own PC

AI dubbing software that runs on your Windows PC. Translate and re-voice video in 10 languages in the original voice, with no minute limits or uploads. DubbingVideo Translation 10 min

Why AI Voices Lose Emotion in Long Audio (And the Fix)

AI voices drift from warm to flat over long audio. Here is why delivery consistency breaks across audiobooks and how Foundry keeps a voice steady inside one rich narration document. AI VoiceTTS 8 min

What GPU Do You Need for Local AI Audio?

Local AI audio needs the right GPU. Here's exactly how much VRAM you need for voice cloning, music generation, and TTS in June 2026, with specific card picks at every budget. Local AIHardware 9 min