Frequently Asked Questions

Live Translate is a Windows desktop AI translation suite distributed on Steam. It translates live audio (games, movies, calls), screenshots, videos, manga / comic / webtoon folders, and entire web pages — all processed locally on your GPU with zero cloud dependency. One single Steam app bundles a real-time speech-to-text engine, an on-device LLM, a Chromium AI browser, an OCR + inpainting pipeline, and an SDXL anime image generator.

Google Translate and DeepL send your text and audio to remote servers over the internet. Live Translate runs everything on your own GPU. There is no API key, no monthly subscription, no usage cap, and no data ever leaves your PC. After installation you can disconnect from the network and every feature continues to work — including live audio translation, video subtitles, manga batch processing, and AI anime art generation.

Because it has no recurring server costs. Cloud translators charge subscriptions to cover their inference servers — Live Translate runs on your hardware, so the price is one-time. The $9.99 also reflects our positioning: a privacy-first tool that anyone can afford, not a niche enterprise product. The price includes lifetime updates, every future Steam build at no extra cost.

Gamers playing Japanese / Korean / Chinese exclusives. Anime, manga, and webtoon readers who want immediate translation. Video editors generating SRTs from raw footage. Language learners watching foreign movies with side-by-side subtitles. Privacy-conscious users who refuse cloud services. And anyone who hates subscriptions.

No. Live Translate runs entirely on your own GPU using bundled open-source AI models. There is no OpenAI, Anthropic, Google, or any other cloud API call anywhere in the app — architecturally enforced. Translation, speech recognition, image generation, OCR, and inpainting all happen on your machine.

Yes, completely. After installation no internet connection is required for any feature. The user can disconnect from the network and every function continues to work: live audio translation, vid2srt, manga batch, anime generation, web translation of already-loaded pages, screenshot inpainting, gallery browsing of local works, hotkeys, overlays, settings. Only Steam Workshop browsing and DLC installs require online Steam — both are explicit user actions.

No. Zero cloud API calls. No OpenAI, Gemini, Anthropic, AssemblyAI, DeepL, Google Translate, or any other network translation service. The architecture forbids it — there is no Live Translate server, no telemetry endpoint, no analytics pixel.

Zero. No analytics, no crash reporters, no anonymous statistics, no "voluntary" data sharing. We do not know how many users we have, what features they use, or which translations they perform. The only network traffic the app generates is what Steam itself does (DLC checks, Workshop browsing, signed-in user info) — and that is Steam, not Live Translate.

In a local SQLite file under your data folder (data/history.db). The cache is in data/cache.db. Web browser chats are in data/web-chat.db. WebView2 user-data is isolated from your system browser profile. Everything stays on your PC. Uninstalling preserves the data folder; you can copy it to back up everything.

Yes. After downloading the Steam package once, no further network access is required for any feature. Workshop and DLC features are optional and gracefully disabled in offline environments. The app launches, models load, and translations run with the network cable unplugged.

A one-time $9.99 USD on Steam. No subscription, no usage limits, no API quotas, no expiry. Steam regional pricing applies in your local currency.

Yes. You buy it once, you own it forever. Every future update — bug fixes, new features, model upgrades, additional languages, performance improvements — is included at no additional cost. Steam handles the update download.

Yes. All app updates are free for the lifetime of the product. The only paid additions are optional DLCs (heavy AI models for users with 12+ / 16+ / 24+ GB VRAM, anime image generators) — these are pure quality upgrades, never required to use the app.

Almost certainly. The version you install today already does an enormous amount, but on some machines it can run slower than expected because the underlying AI technology is still young. That technology is changing rapidly, and Live Translate is changing with it — every Steam update brings faster engines, smaller models, and sharper output, all at no additional cost. If you bought on day one, every future improvement is yours for free. The earlier you join, the more upgrades you receive on the same one-time purchase.

Standard Steam refund policy applies: refundable within 14 days of purchase if you have played less than 2 hours.

On the Steam store: store.steampowered.com/app/3812790/Live_Translate/. The Steam client downloads the app to your library and handles all updates and DLC installs.

The user interface is fully localized in 29 languages — the complete Steam-supported set. Translation between language pairs covers 140+ languages via Gemma 4. Speech-to-text supports 28+ languages via Vosk with additional coverage from Whisper (99 languages), Moonshine (English), and Kyutai (English + French).

Speech recognition covers 28+ commonly spoken languages out of the box (including Arabic, Chinese, Czech, Danish, Dutch, English, Farsi, Finnish, French, German, Greek, Hindi, Italian, Japanese, Kazakh, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Tagalog, Turkish, Ukrainian, Uzbek, and Vietnamese). On top of that, an additional GPU-based engine covers 99 source languages — including most low-resource ones — for users with 8 GB VRAM or more.

Yes — webtoons are a first-class use case. The Web Translator tab's auto-translate-on-scroll is purpose-built for vertical Korean strips. It pauses on scroll, captures the visible chunk, and translates as you read downward. The Story Changer pipeline handles arbitrary page heights with no length limit. A dedicated "Webtoon (Korean)" art preset reproduces the full-color digital aesthetic for image generation.

Yes. Manga right-to-left reading order is detected automatically in batch and screenshot inpaint modes. Arabic, Hebrew, and other RTL languages are supported as both source and target. Speech bubble shape preservation reflows text within original bubble contours regardless of script direction.

Recommended: Windows 10 (1809+) or Windows 11 64-bit, modern CPU, 16 GB RAM, and a GPU with 8 GB of VRAM. 50 GB of free disk space for the base install. 8 GB GPU VRAM is the sweet spot for the full live-translation, video subtitle, and batch manga experience. Bigger cards unlock heavier features such as image generation and longer Story Changer projects.

It launches and translates, but with reduced quality and noticeably slower throughput. Casual live translation, screenshot mode, and small batches work on a CPU-only fallback. For a smooth experience and access to the heavier modules (image generation, full-quality video subtitles), a GPU with 8 GB of VRAM is strongly recommended.

Yes — but not equally. NVIDIA is the recommended path and gives the fastest performance. AMD is fully supported and runs every module, but tends to be noticeably slower than an equivalent NVIDIA card on most workloads. Intel works for casual use; for heavy modules (image generation, full-quality video subtitles) we recommend an 8 GB NVIDIA or AMD GPU.

The open-source AI engines that Live Translate bundles have years of optimization for NVIDIA's CUDA stack — that's where the entire AI ecosystem grew up. AMD support uses the Vulkan compute path, which is mature and works on every module, but for the same VRAM tier you typically see slower live translation throughput, slower video transcription, and slower image generation on AMD. The gap is roughly 1.5×–3× depending on the workload. For pure offline translation work, an NVIDIA card with 8 GB of VRAM will feel snappier than an AMD card with 12 GB of VRAM. AMD is improving fast — if you already own an AMD card, Live Translate works; if you are buying new specifically for this app, NVIDIA is the better pick.

Only if you fall below 8 GB of VRAM. The 8 GB tier covers the entire feature set comfortably — live translation, video subtitles, manga batch, the AI browser. Above 8 GB you unlock heavier features (image generation, longer Story Changer projects, parallel modules), but the experience at 8 GB is already very usable. If your current card has 4–6 GB you can still run the app, just with reduced quality and slower throughput. Upgrading from a 12 GB NVIDIA to a 16 GB NVIDIA is rarely worth it for Live Translate alone — the marginal gain is small unless you do heavy image generation.

Modern laptop GPUs with 8 GB of VRAM (e.g. mobile RTX 4060 / 4070, Radeon 7700M with 8 GB) run the full feature set well. Power throttling and thermal throttling will affect long batch jobs more than desktop GPUs, but for live translation (continuous low-load workload) and short batches the experience is essentially the same as on desktop. We recommend plugging in to AC for any heavy session — the GPU pulls full power and battery life takes a serious hit during AI inference.

No, Live Translate runs on a single GPU. If you have two cards in one machine, the app picks one (the most powerful by default, configurable in Settings). Multi-GPU split inference adds significant complexity for marginal gain on consumer workloads, so we kept it simple — one card, full feature set, less can-go-wrong.

Yes. The app you install today is a snapshot of an evolving stack. Every Steam update brings smaller, faster AI models, better quantization, and new hardware optimizations. A workload that takes 4 seconds today may take 1.5 seconds on the same hardware in a year. All of these improvements ship for free as part of your one-time $9.99 purchase — the lifetime updates promise is real and ongoing.

Cloud translation services have to do a network round-trip for every request — typically 100–500 ms before any work even starts. Offline AI on your own GPU has zero network latency, so the moment you press a hotkey or stop talking, inference begins. After warmup (model loaded into VRAM), live translation responses feel essentially instant. The only thing slower than cloud is the very first request after the app launches — once the model is resident, every subsequent translation is faster.

About 50 GB of free disk space for the base install — that covers the app, runtime, all bundled translation and speech models, language packs, and context presets. Optional Steam DLCs (image generators, advanced translation models) add 7–20 GB each. Plan for 60–80 GB free if you intend to install the heavier DLCs.

Currently Windows 10/11 x64 only. Mac and Linux are not on the published roadmap. The app is built on .NET 8 + WPF, which is Windows-native; porting would require substantial rework.

Yes — Live Translate ships every AI model bundled in the Steam download, so the base install is around 50 GB. That avoids extra downloads at runtime, lets the app work offline immediately after install, and keeps the model files versioned together with the app. Optional DLCs (image generators, advanced translation models) add 7–20 GB each on top.

Yes — that's the primary use case. The transparent overlay window stays on top of any game, you can drag it anywhere, click-through optionally so it never intercepts mouse input. Pick "system loopback" as the audio source and Live Translate translates whatever your PC is playing — including Discord, voice chat, in-game cinematics, all simultaneously. Default hotkeys (Ctrl+Alt+S to toggle, Ctrl+Alt+H to hide overlay) are fully rebindable.

There is no hard length limit. We have run multi-hour movies, 12-hour podcast archives, and full anime seasons through the same pipeline. The bottleneck is disk space (transcription cache stores intermediate state) and your patience — a 2-hour video on the recommended 8 GB GPU finishes in roughly 15–25 minutes for both transcription and translation combined. Multi-language output from one transcription is the killer feature here: transcribe once, generate Turkish + English + Japanese SRTs in seconds each.

Yes, very much so. Watch a Japanese movie with the live translation overlay running — you see the Japanese audio AND your native-language translation in real time. Read foreign blogs / Wikipedia / news articles with the AI Browser's auto-translate-on-scroll. Generate side-by-side SRT files (original + translation) for video lessons. The 102 game / movie context presets help with proper noun pronunciation and idiom handling. Privacy-conscious learners appreciate that every conversation, every page, every audio file stays on their PC — the app cannot phone home.

Yes, three different ways: (1) screenshot mode for ad-hoc pages with AI inpainting that removes original text and renders translation in the original layout; (2) batch mode for entire folders of pages with session resumption; (3) Story Changer mode that rewrites the entire story with new characters / setting / theme and stylizes every page — producing a new fan-comic.

Yes, via the Anime Creator tab. Three modes: text-to-image (T2I) from a prompt, image-to-image (I2I) photo-to-anime stylization, and Story Changer multi-page rewrite. 69 art presets across 11 categories plus full LoRA support. Image generation requires installing one of the optional Image DLCs from Steam (Anime XL or Anime SM).

The Vid2Srt tab takes any video file (MP4 / MKV / AVI / MOV / WebM / WMV / FLV / M4V) and produces a polished SRT file in any of 140+ target languages using a two-pass LLM pipeline (per-segment translation, then full-script coherence review). Multi-video queue, multi-language output from one transcription, and hard-burn re-encoding with NVENC / AMF / QuickSync hardware acceleration are all built in.

A built-in WebView2 (Chromium) browser inside the app with three superpowers: (1) auto-translate-on-scroll for any web page including webtoons; (2) per-URL multi-turn AI chat that remembers conversations when you revisit the page; and (3) an autonomous AI Browse agent that can click, type, and navigate to complete tasks like "find me the cheapest flight" or "summarize this article in Turkish". WebView2 user-data is isolated from your system browser.

Right-click any output in the My Works tab → "Share to Community". Live Translate auto-fills title from prompt, builds a rich description, includes generation parameters, and uploads to Steam Workshop. Multi-select batch upload is supported for sharing multiple items in one click.

All user data lives under a single local data folder (translation history, chat memory, generated images, video subtitles, project files, custom bubbles). Copy this folder to back up everything — translations, conversations, generated artwork, every project you ever started. Restoring is the reverse: copy the folder back and the app picks up exactly where you left off. Uninstalling the app preserves the folder; reinstalling reads it on first launch.

Live Translate ships with multiple bundled open-source models — large language models for translation, several speech-recognition engines, image-generation models for the DLCs, and dedicated OCR and inpainting models for manga. All models are commercially-permissive open licenses (Apache 2.0, MIT, CC-BY-4.0, Fair AI Public License). The full attribution list is shipped with the app and is also visible on the About page.

The Live Translate application itself is proprietary (Steam commercial license). All bundled AI models use commercially-permissive open licenses: Apache 2.0, MIT, CC-BY-4.0, Fair AI Public License. See the bundled THIRD-PARTY-NOTICES.txt for full attribution.

Updates ship through Steam whenever new features, model upgrades, or significant fixes are ready. There is no fixed cadence — quality and stability come first. All updates are free for the lifetime of the product.

Email [email protected] or use the Steam Discussions on the Live Translate store page. Bug reports with reproduction steps and a copy of the relevant log file (under data/logs/, 7-day rotated) are most useful.