Question 1

What is VidNo and how does it work?

Accepted Answer

VidNo is an AI-powered video production pipeline built specifically for developers. You run one bash command and VidNo handles everything from screen recording analysis to YouTube upload — scripted, narrated, edited, thumbnailed, published. No video editing skills required. No manual voiceover. No timeline scrubbing. No copying titles into YouTube Studio. Nothing manual.

Here is how the pipeline works from start to finish. You record your screen while coding, using any tool you prefer — OBS, your OS's built-in recorder, even Loom. When you are done, you run a single command and VidNo takes it from there.

VidNo's first pass is visual analysis. It extracts frames from your recording at regular intervals and runs OCR to read everything on screen — your code editor, terminal output, browser tabs, file trees. It also detects git diffs by analyzing what changed between frames, giving it a precise understanding of what you actually built during the session.

Next, all that extracted context gets sent as text to Claude (Anthropic's AI) via API. Claude generates a natural, developer-friendly script that explains what you did, why you did it, and walks through the key decisions. The script reads like you wrote it yourself, not like a generic tutorial.

Then VidNo handles the voice. Using MOSS TTS running locally on your GPU, it synthesizes speech from the script using a voice model trained on a 60-second sample of your actual voice. The result sounds like you narrating your own work — matching your cadence, rhythm, and emphasis patterns.

VidNo composites everything together — intelligent cuts to remove dead time, transitions between logical sections, voiceover synced to the relevant screen content — and renders the final output. You get four versions from every recording: a full-length tutorial, a condensed recap, a short highlight reel, and a vertical YouTube Short.

After rendering, VidNo generates a custom thumbnail for each video. Then it uploads everything directly to YouTube via API — setting the title, description, tags, chapters, thumbnail, and schedule. Your videos go from screen recording to published on YouTube without you touching a browser.

The entire process runs locally on your machine. Your screen recordings and code never leave your computer — only the text extracted for script generation touches the cloud via Claude's API, and the final upload to YouTube via the YouTube Data API.

Question 2

Who is VidNo built for?

Accepted Answer

VidNo is built for anyone who writes code and wants to publish video content without learning video editing. The core audience is developers who understand that YouTube is the most powerful distribution channel for technical content but cannot justify spending hours editing footage every week.

Indie hackers and solo founders are the most obvious fit. You are already building in public, shipping features, and writing changelogs. VidNo turns your daily coding sessions into polished build-in-public videos that drive traffic to your product. Record your screen while you work, and VidNo handles the rest. Your audience gets to watch you build the product they use.

Freelance developers benefit by creating case study videos from client work (with permission). A portfolio full of narrated walkthroughs showing real problem-solving is significantly more convincing than a static list of past projects. VidNo produces those walkthroughs without adding editing overhead to your workflow.

Open source maintainers can use VidNo to create contribution guides, feature walkthroughs, and release videos. Good documentation is the number one driver of OSS adoption, and video walkthroughs dramatically reduce the barrier to understanding a codebase.

Dev agencies use VidNo to produce internal training content and client-facing progress updates. Recording a sprint demo session and running it through VidNo generates a professional update video without dedicating anyone to video production.

Tech educators and course creators benefit from VidNo's ability to handle the tedious parts — cutting dead time, generating narration, producing multiple formats. You focus on teaching, VidNo focuses on production.

The common thread across all these users is the same: they have valuable coding content trapped in screen recordings and no time or desire to manually edit it into watchable videos. VidNo bridges that gap entirely.

Question 3

What makes VidNo different from Descript or Gling?

Accepted Answer

The fundamental difference is that VidNo understands code. Descript, Gling, Kapwing, and every other AI video editor on the market treat your screen recording as a stream of pixels. They can detect silence, identify filler words, and make basic cuts — but they have zero understanding of what is happening on screen.

VidNo reads your screen. It runs OCR on editor content, parses terminal output, detects git diffs between frames, identifies which files changed and how. When it generates a script, it references specific functions, explains architectural decisions, and walks through the logic of your changes. Descript cannot do this because it does not know the difference between a React component and a spreadsheet.

The second major difference is that VidNo handles the ENTIRE pipeline — not just video creation but also thumbnail generation, YouTube Shorts creation, and direct upload to YouTube with full metadata (title, description, tags, chapters, thumbnail, scheduling). Other tools stop at rendering a file. VidNo stops when your video is live on YouTube. One command, zero browser tabs.

Third, VidNo runs locally. Descript and Gling upload your recordings to their cloud servers for processing. For developers working on proprietary codebases, client projects, or anything under NDA, that is a non-starter. VidNo processes everything on your machine. The only external calls are text sent to Claude's API for script generation and the final upload to YouTube — no video frames, no screenshots, no audio leaves your computer during processing.

Fourth, VidNo produces four distinct outputs from every recording. Most editors help you clean up one video. VidNo generates a full tutorial, a quick recap, a short-form highlight reel, and a vertical YouTube Short — each optimized for different platforms and audience segments, each uploaded automatically. That is four pieces of content from one coding session.

Fifth, the voice synthesis is local. Descript's voice cloning runs on their servers. VidNo's MOSS TTS runs on your GPU. Your voice model stays on your hardware.

Finally, VidNo is built for a developer workflow. It is a CLI tool, not a drag-and-drop timeline editor. You run a command, review the generated script, and let it render and upload. No timeline, no tracks, no keyframes. If your workflow lives in the terminal, VidNo fits naturally into it.

Question 4

How many videos does VidNo produce from one recording?

Accepted Answer

Every screen recording you process through VidNo generates four distinct output videos, each optimized for a different purpose and audience. All four are uploaded automatically to YouTube via API — with titles, descriptions, tags, chapters, thumbnails, and scheduling handled by VidNo.

The first is the full tutorial. This is a comprehensive, long-form video that walks through your entire coding session from start to finish. VidNo's script covers every significant change, explains the reasoning behind architectural decisions, and provides the kind of context that makes a tutorial genuinely useful. These typically run 10-20 minutes depending on your recording length and are ideal for YouTube's long-form algorithm, which rewards watch time and session duration.

The second is the quick recap. This is a condensed 3-5 minute version that hits the highlights without the deep-dive explanations. VidNo identifies the most important moments — the key feature implementations, the critical bug fixes, the architectural pivots — and builds a focused narrative around just those moments. Quick recaps work well for audiences who already understand the technology and want to see what you built without sitting through the full walkthrough. They are also effective for social media sharing where attention spans are shorter.

The third is the highlight reel. This is a sub-60-second vertical video optimized for TikTok and Instagram Reels. VidNo selects the single most visually interesting or impressive moment from your session and builds a punchy, fast-paced clip around it. The format follows short-form best practices — immediate hook, quick payoff, clear call to action pointing viewers to the full tutorial.

The fourth is the YouTube Short. VidNo generates a dedicated vertical Short with its own script, pacing, and hook optimized specifically for the YouTube Shorts algorithm. This is not just a cropped version of the highlight reel — it is a separate output written for YouTube's discovery feed with platform-specific formatting.

All four videos use the same voice model and visual style, so your channel maintains a consistent feel. The scripts are generated independently — the short versions are not just trimmed copies of the long version. Each is written specifically for its format and audience expectations. After rendering, VidNo uploads everything to YouTube automatically — you never open YouTube Studio.

This four-output approach means a single coding session can fuel your content calendar for an entire week across multiple platforms, fully published without manual effort.

Question 5

What programming languages does VidNo support?

Accepted Answer

VidNo is completely language-agnostic. It works with any programming language, framework, or tool that you can display on your screen. There is no configuration, no plugin system, and no language-specific setup required.

This is possible because VidNo does not parse your code through a compiler or AST. Instead, it reads your screen content through OCR — the same way a human would watch your recording. It extracts text from your code editor, terminal, browser, and any other visible window. It also detects git diffs by comparing frames over time to identify what changed in your files.

That extracted text gets sent to Claude for script generation. Claude has deep knowledge of virtually every programming language in active use, so it can intelligently explain Python, JavaScript, TypeScript, Go, Rust, Java, C++, C#, Ruby, PHP, Swift, Kotlin, Elixir, Haskell, Scala, Dart, Lua, R, SQL, and dozens more. It understands frameworks too — React, Next.js, Django, Rails, Spring Boot, FastAPI, Express, SvelteKit, you name it.

Beyond traditional programming, VidNo works just as well for infrastructure and DevOps content. Recording yourself writing Terraform configs, Kubernetes manifests, Docker Compose files, CI/CD pipelines, or shell scripts produces equally strong results. The OCR captures the content and Claude understands the context.

VidNo also handles mixed-language sessions naturally. If your coding session involves writing a Go backend, a TypeScript frontend, and some SQL migrations, VidNo tracks all of it and produces a coherent script that covers each layer.

The only practical limitation is screen readability. If your font size is extremely small or your recording resolution is low, OCR accuracy drops. We recommend at least a 14px font and 1080p recording resolution for reliable text extraction. Beyond that, any language works out of the box.

Question 6

What GPU do I need to run VidNo?

Accepted Answer

VidNo requires an NVIDIA GPU with CUDA support for voice synthesis and video rendering. AMD and Intel GPUs are not currently supported because the MOSS TTS voice cloning model depends on CUDA-specific operations.

The minimum supported GPU is the NVIDIA RTX 3060 with 12GB VRAM. At this tier, VidNo runs correctly but processing is slower. Voice synthesis for a 15-minute script takes roughly 8-10 minutes, and full video rendering adds another 15-20 minutes on top of that. Usable for occasional use, but you will feel the wait on longer recordings.

The recommended GPU is the RTX 4070 Ti Super or above with 16GB VRAM. This is the sweet spot where processing times become comfortable — roughly a 1:3 ratio between output length and processing time. A 15-minute tutorial renders in about 5 minutes.

The optimal GPU is the RTX 4090 with 24GB VRAM. At this tier, VidNo flies. Voice synthesis runs in near real-time, video rendering is fast, and you can process longer recordings (60+ minutes) without VRAM pressure.

Here is a rough benchmark table for rendering a 30-minute recording into three output videos:

RTX 3060 12GB — approximately 35-45 minutes total processing.
RTX 3070 Ti 8GB — not recommended, VRAM too low for reliable voice cloning.
RTX 4060 Ti 16GB — approximately 20-25 minutes total processing.
RTX 4070 Ti Super 16GB — approximately 12-15 minutes total processing.
RTX 4080 Super 16GB — approximately 10-12 minutes total processing.
RTX 4090 24GB — approximately 8-10 minutes total processing.

VRAM is the critical bottleneck, not raw compute speed. The voice model needs to stay resident in VRAM during synthesis, and if it gets swapped out, quality degrades significantly. Close other GPU-intensive applications (games, other AI tools, GPU-accelerated browsers) before running VidNo.

Make sure your NVIDIA drivers are up to date. Run nvidia-smi to verify your GPU is detected and check available VRAM before your first VidNo run.

Question 7

What are VidNo's full system requirements?

Accepted Answer

VidNo has specific hardware and software requirements because it runs AI inference, video processing, and voice synthesis locally on your machine. Here is the full breakdown.

Operating system: Linux is required. Ubuntu 22.04 LTS or newer is recommended and is the primary development and testing target. Other Debian-based distributions (Linux Mint, Pop!_OS, Debian 12) work well. Fedora and Arch are community-supported but not officially tested. macOS support is planned but not yet available. Windows is not supported natively — WSL2 with NVIDIA GPU passthrough can work but is not officially supported.

CPU: Any modern multi-core processor. 8 cores or more recommended. The CPU handles OCR, frame extraction, and video compositing. An Intel i7/i9 or AMD Ryzen 7/9 from the last three generations will be comfortable.

RAM: 32GB minimum, 64GB recommended. VidNo loads video frames into memory during analysis, and the OCR pipeline is memory-intensive when processing high-resolution recordings. 16GB will cause out-of-memory errors on recordings longer than 20 minutes.

GPU: NVIDIA with 12GB+ VRAM and CUDA support. See the GPU requirements FAQ for specific model benchmarks. AMD GPUs are not supported.

Storage: 100GB of free disk space minimum. Screen recordings are large files, and VidNo creates intermediate artifacts during processing (extracted frames, audio files, composited segments) before cleaning up. An SSD is strongly recommended — HDD speeds will bottleneck the frame extraction phase significantly.

Software dependencies: Node.js 20 or newer, FFmpeg (installed via apt), NVIDIA drivers with CUDA toolkit, and a Claude API key from Anthropic. The VidNo installer checks for all dependencies and guides you through installing anything missing.

Network: An internet connection is required only for the Claude API call during script generation. All other processing is fully offline. Typical bandwidth usage per video is minimal — a few kilobytes of text sent and received.

Question 8

How do I install VidNo?

Accepted Answer

Installing VidNo takes about 10 minutes. The process has four steps: download, run the installer, configure your API key, and record your voice sample.

Step one: download. Grab the latest release from the VidNo downloads page. You will get a compressed archive — vidno-latest.tar.gz. Extract it to wherever you want VidNo to live. Most users put it in their home directory or /opt.

Step two: run the installer. Open a terminal, navigate to the extracted directory, and run ./install.sh. This script checks your system for all required dependencies — Node.js 20+, FFmpeg, NVIDIA drivers, CUDA toolkit — and installs anything missing (with your confirmation). It sets up the VidNo CLI so you can run vidno commands from anywhere in your terminal.

Step three: configure your Claude API key. Run vidno config and paste your Claude API key when prompted. If you do not have one yet, sign up at console.anthropic.com, navigate to API Keys, and create a new key. VidNo stores the key in ~/.vidno/config.json on your local machine. The key is never sent anywhere except Anthropic's API endpoint during script generation.

Step four: record your voice sample. Run vidno voice-setup and follow the prompts. You will read a provided paragraph out loud for about 60 seconds. Speak naturally — use your normal talking voice, not a performance voice. The quality of this sample directly affects how good your AI narration sounds. Record in a quiet environment, speak at a comfortable pace, and avoid background noise. VidNo processes the sample on your GPU to create a local voice model, which takes 2-5 minutes depending on your hardware.

After these four steps, you are ready to process your first recording. Run vidno process your-recording.mp4 and VidNo handles everything from there.

If anything fails during installation, the installer outputs specific error messages with suggested fixes. The most common issues are outdated NVIDIA drivers and missing CUDA toolkit, both of which the installer can resolve automatically on Ubuntu systems.

Question 9

Does VidNo work on Mac or Windows?

Accepted Answer

VidNo currently runs on Linux only. Ubuntu 22.04 LTS and newer is the primary supported platform. Other Debian-based distributions like Linux Mint, Pop!_OS, and Debian 12 are also well-tested and work reliably.

The Linux-first decision is driven by two technical factors. First, NVIDIA's CUDA toolkit has the best support and most consistent behavior on Linux. The MOSS TTS voice cloning model requires CUDA for inference, and Linux provides the most stable and performant CUDA environment. Second, VidNo uses FFmpeg extensively for video processing, and Linux's FFmpeg ecosystem (hardware-accelerated encoding, codec support, pipeline integration) is more mature and reliable than on other platforms.

For macOS: support is planned and actively being worked on. The main blocker is Apple Silicon GPU support for the voice synthesis model. MOSS TTS was built for CUDA, and porting it to Metal (Apple's GPU framework) requires non-trivial work. Mac users with NVIDIA eGPUs could theoretically run VidNo today, but this is not tested or supported. When macOS support ships, it will target M1 Pro and above with at least 32GB unified memory.

For Windows: native Windows support is not on the roadmap. However, WSL2 (Windows Subsystem for Linux) with NVIDIA GPU passthrough does work for some users. You need Windows 11, WSL2 with an Ubuntu 22.04 distribution, and NVIDIA's WSL2 CUDA drivers installed. This setup mirrors a native Linux environment closely enough that VidNo functions correctly. That said, it is not officially supported — if you run into WSL2-specific issues, we may not be able to help debug them.

If you are a developer who primarily uses macOS or Windows, the most reliable path today is a dedicated Linux machine or a dual-boot setup. Many VidNo users keep a Linux partition or a secondary machine specifically for content production. A used desktop with an RTX 3060 runs VidNo comfortably and can be set up for under $500.

Question 10

How do I get a Claude API key for VidNo?

Accepted Answer

VidNo uses Claude (by Anthropic) to generate video scripts from the code context it extracts from your screen recordings. You need your own API key, which takes about two minutes to set up.

Go to console.anthropic.com and create an account if you do not already have one. Once logged in, navigate to the API Keys section in the left sidebar. Click "Create Key," give it a name like "vidno" so you can identify it later, and copy the key. It starts with sk-ant- and is a long string of characters.

Back in your terminal, run vidno config and paste the key when prompted. VidNo stores it in ~/.vidno/config.json on your machine. The key is only used when VidNo sends extracted text to Claude's API for script generation. No video, audio, or screenshot data is ever sent — only the text content that OCR pulled from your screen and any git diff summaries.

Regarding cost: Claude API usage for VidNo is remarkably cheap. A typical 30-minute coding session produces about 3,000-5,000 tokens of extracted context, and Claude generates a 2,000-3,000 token script in response. At current Claude API pricing, that works out to roughly $0.003 per video — about a third of a cent. Even if you produce 10 videos per week, your monthly Claude API bill would be under $0.15.

Anthropic bills based on usage with no minimum commitment. You load credits into your account and they are drawn down as you make API calls. A $5 initial deposit will last you through hundreds of VidNo videos.

One important note: VidNo uses the Claude API, not the Claude chat interface. They are separate products with separate billing. A Claude Pro chat subscription ($20/month) does not include API access. You need API credits loaded at console.anthropic.com specifically.

Keep your API key private. Do not commit it to git repositories or share it publicly. If you believe your key has been compromised, revoke it immediately in the Anthropic console and generate a new one.

Question 11

How does voice cloning work in VidNo?

Accepted Answer

VidNo uses MOSS TTS, a local text-to-speech model that runs entirely on your GPU, to clone your voice and generate narration for your videos. The process starts with a 60-second voice sample and produces a personal voice model that captures not just how you sound, but how you speak.

During setup, you run vidno voice-setup and read a provided paragraph out loud. The paragraph is specifically designed to capture a wide range of phonemes, intonation patterns, and speech rhythms. Speak naturally — the model works best when it learns your actual conversational tone rather than a stiff "recording voice."

MOSS TTS analyzes your sample to extract several layers of vocal characteristics. The obvious ones are pitch, timbre, and tone — what makes your voice sound like you rather than someone else. But it also captures speaking rhythm: where you tend to pause, how you handle emphasis, your pace when explaining something versus when you are running through setup steps. This rhythm layer is what separates VidNo's voice output from generic TTS that sounds robotic even when it mimics your timbre.

The resulting voice model is a set of weights stored locally on your machine at ~/.vidno/voices/. The model file is typically 200-400MB depending on your GPU's precision settings. It never gets uploaded anywhere.

When VidNo generates narration for a video, it feeds the script text through your voice model on your GPU. The synthesis happens in chunks — sentence by sentence — so VidNo can apply appropriate pauses between sections and adjust pacing based on the content type. Code explanations get slightly slower pacing; transitional phrases move quicker.

Voice quality depends on three factors: GPU capability (more VRAM allows higher-fidelity synthesis), sample quality (quiet room, clear speech, no background noise), and script quality (natural-sounding sentences produce more natural-sounding speech). On an RTX 4070 or above with a clean voice sample, most listeners cannot distinguish the AI narration from a real human recording.

Question 12

How good is the AI voice quality?

Accepted Answer

In blind listening tests conducted during development, 73% of participants could not reliably distinguish VidNo's AI-generated narration from a human recording of the same script. That number climbs to 81% when the voice sample was recorded in professional conditions (quiet room, decent microphone, natural speech).

The quality you get depends on three variables that are entirely within your control.

First is your GPU. Voice synthesis fidelity scales directly with available VRAM and compute power. An RTX 3060 produces good results — clearly your voice, natural cadence, no robotic artifacts. An RTX 4070 Ti or higher produces excellent results where the AI voice is essentially indistinguishable from a real recording in casual listening. The difference is subtle but noticeable in direct A/B comparisons: higher-end GPUs produce more nuanced emphasis and more natural breath patterns.

Second is your voice sample quality. This is the single biggest factor. A 60-second sample recorded on a laptop microphone in a room with AC humming will produce mediocre results regardless of GPU. A sample recorded with a decent USB microphone (even a $50 one) in a quiet room with natural, relaxed speech will produce dramatically better output. The model can only learn what you give it.

Third is the script itself. MOSS TTS handles conversational, natural-sounding text much better than stiff or overly formal writing. Claude's script generation is tuned to produce natural developer-friendly language, which helps. But if you manually edit the script into something that sounds unnatural when read aloud, the voice synthesis will sound unnatural too.

Compared to cloud-based TTS services like ElevenLabs or Play.ht, VidNo's local MOSS TTS trades a small amount of raw voice fidelity for complete privacy and zero per-minute costs. Cloud services are marginally better in pure audio quality on certain voice types, but VidNo's output is well above the threshold where viewers notice or care — especially for technical content where the information matters more than broadcast-quality audio.

Question 13

Can I use my own voice or do I have to use AI?

Accepted Answer

Both options are fully supported. VidNo's AI voice cloning is a feature, not a requirement. You can choose whichever approach fits your workflow and preferences.

Option one: use the AI-cloned voice. This is the default workflow and what most users choose. VidNo generates a script from your coding session, synthesizes speech using your local voice model, and renders the video with AI narration. The advantage is speed — you never have to record a voiceover, and you can produce videos from coding sessions after the fact without being near a microphone.

Option two: record your own voiceover. If you prefer narrating yourself, VidNo supports a manual voiceover workflow. The process is: run VidNo in script-only mode (vidno process --script-only recording.mp4), review and edit the generated script, record yourself reading it, then run VidNo in render mode with your audio file (vidno render recording.mp4 --audio your-voiceover.wav). VidNo handles the editing, cuts, and compositing using your real voice instead of the AI clone.

Option three: hybrid. Some users let VidNo generate the AI voice for the initial render, review the output, and then re-record only the sections where they want a more personal touch. You can mix AI and human voiceover within the same video by providing timestamps for which sections use which audio source.

There is no quality penalty for choosing any option. The video editing, smart cuts, transitions, and compositing work identically regardless of the audio source.

One common use case for the own-voice option is developers who already have an established YouTube channel with a recognizable voice. Their audience knows how they sound, and switching to AI narration would be jarring. These users love VidNo for the script generation and editing pipeline while keeping their authentic voice. The AI voice is there for when speed matters more than that personal touch — batch-producing content, covering less important updates, or creating secondary channel content.

Question 14

Can I create multiple voice profiles?

Accepted Answer

Yes, VidNo supports multiple voice profiles. The number of profiles available depends on your plan.

On the free tier, you get one voice profile. This covers the standard use case — your voice, trained from your 60-second sample, used for all your video narration. For most individual developers, one profile is all you need.

On the Pro plan, you can create up to five voice profiles. This opens up several useful workflows. Some creators maintain different speaking styles for different content types — a more energetic, faster-paced profile for short-form highlight reels, and a calmer, more deliberate profile for long-form deep-dive tutorials. Same voice, different energy. You record separate 60-second samples for each style, and VidNo builds distinct models.

On the Team plan, each of the five team members gets their own set of voice profiles (up to five per member). This is designed for agencies and development teams where multiple people contribute coding content under the same channel or brand. Each person records their own voice sample, and VidNo produces videos narrated in the correct voice for whoever did the coding.

Creating a new voice profile is the same process as the initial setup. Run vidno voice-setup --profile "profile-name" and record your sample. Switch between profiles when processing videos with the --voice flag: vidno process recording.mp4 --voice "tutorial-style".

Voice profiles are stored locally in ~/.vidno/voices/ as separate model files. Each model is 200-400MB. You can back them up, move them between machines, or delete them at any time. Exporting voice profiles is supported on all paid plans, so if you cancel your subscription, you keep your trained voice models.

A practical note: the voice sample environment matters more than the number of profiles. One excellent profile recorded with a good microphone in a quiet room will outperform five profiles recorded with laptop mics in noisy spaces. Quality over quantity.

Question 15

What quality are the output videos?

Accepted Answer

VidNo produces professional-grade video output that is uploaded directly to YouTube via API — no manual upload step, no copying metadata into YouTube Studio. The rendered files are optimized for YouTube's encoding pipeline so they retain maximum quality after re-compression.

The default output resolution is 1920x1080 (1080p) at 30 frames per second. This matches the resolution of most developer screen recordings and produces crisp, readable code on screen. Text clarity is a priority — VidNo's rendering pipeline is tuned to ensure that code, terminal output, and UI elements remain sharp and legible in the final output, even after YouTube's compression.

On the Pro plan, you unlock 4K output (3840x2160) and 60fps rendering. 4K is particularly valuable if your screen recordings are captured at 4K or if you use high-DPI displays. The extra resolution ensures that dense code with small font sizes remains readable even on smaller playback screens. 60fps is useful for recordings that involve smooth scrolling, animations, or browser interactions where 30fps can look stuttery.

The video codec is H.264 (AVC) with a high bitrate profile. VidNo defaults to a bitrate of 12 Mbps for 1080p and 40 Mbps for 4K, which is well above YouTube's recommended upload bitrate. This means your video retains maximum quality through YouTube's re-encoding process. The audio codec is AAC at 320kbps, which preserves the full fidelity of both the AI voice and any system audio captured during recording.

Container format is MP4 for maximum compatibility. Every platform accepts MP4 without conversion.

For the short-form outputs (highlight reel and YouTube Short), VidNo automatically renders in 9:16 vertical format at 1080x1920, optimized for YouTube Shorts and other vertical video platforms. It intelligently crops the most relevant portion of your screen rather than just letterboxing the horizontal recording.

File sizes for the full tutorial output typically range from 500MB to 2GB depending on length and resolution. The quick recap, highlight reel, and YouTube Short are proportionally smaller. All output files land in ~/vidno/output/ organized by date and session name, and are uploaded to YouTube automatically as part of the pipeline.

Question 16

How long does it take to process a video?

Accepted Answer

Processing time depends on three factors: your GPU, the length of your recording, and the complexity of what is on screen. The general rule of thumb is a 1:3 ratio on a recommended GPU — a 30-minute recording takes about 10 minutes to process into three output videos.

Here is a more detailed breakdown of where time is spent in the pipeline:

Frame extraction and OCR takes about 10-15% of total processing time. VidNo extracts frames at regular intervals and runs optical character recognition on each one to read code, terminal output, and UI text. Longer recordings with more text-dense screens (multiple editor splits, small fonts, lots of terminal output) take proportionally longer in this phase.

Script generation via Claude API takes 30-60 seconds regardless of recording length. This is the fastest step and the only one that requires an internet connection. Claude processes the extracted context and generates scripts for all four output videos in a single API call.

Voice synthesis takes about 35-45% of total processing time. This is the most GPU-intensive phase. MOSS TTS generates speech for each of the four scripts sequentially. The full tutorial script is the longest and takes the most time. On an RTX 4090, voice synthesis runs at roughly 2x real-time speed — a 15-minute script synthesizes in about 7-8 minutes.

Video compositing and rendering takes about 25-30% of total processing time. VidNo makes intelligent cuts, adds transitions, syncs voiceover to screen content, and renders the final MP4 files for all four outputs. This phase uses both GPU (for hardware-accelerated encoding) and CPU (for compositing logic).

Thumbnail generation takes about 5% of total processing time. VidNo creates custom thumbnails for each video using key frames from the recording combined with text overlays optimized for YouTube click-through rates.

YouTube upload takes about 5-10% of total processing time depending on your internet connection speed and the total file sizes. VidNo uploads all four videos via the YouTube Data API, setting titles, descriptions, tags, chapters, thumbnails, and scheduling for each. This runs at the end of the pipeline so your local processing is not bottlenecked by upload bandwidth.

Approximate total processing times for a 30-minute recording (excluding upload):

RTX 3060 12GB — 35-45 minutes.
RTX 4060 Ti 16GB — 20-25 minutes.
RTX 4070 Ti Super 16GB — 12-15 minutes.
RTX 4080 Super 16GB — 10-12 minutes.
RTX 4090 24GB — 8-10 minutes.

Add upload time on top of these estimates — typically 5-15 minutes depending on your connection speed and total output size. You can continue using your computer during processing, but avoid GPU-heavy tasks. VidNo runs at lower priority to minimize interference with your normal workflow.

Question 17

Can I edit the script before VidNo renders the video?

Accepted Answer

Yes, script review and editing is a built-in step in VidNo's pipeline. You always have the option to review, modify, or completely rewrite the AI-generated script before any voice synthesis or rendering happens.

When you run vidno process recording.mp4, VidNo extracts context from your recording and generates scripts via Claude. Before proceeding to voice synthesis, it pauses and presents you with the generated scripts in your terminal. You see the full tutorial script, the quick recap script, and the highlight reel script, clearly separated and labeled.

At this point you have several options. You can approve all scripts as-is by pressing Enter, and VidNo continues to voice synthesis, rendering, thumbnail generation, and YouTube upload — all automatically, no further input needed. You can open any script in your preferred text editor for modifications by pressing e. Or you can reject a script entirely and have Claude regenerate it with additional guidance by pressing r and typing your feedback.

The editing step is where many users add their personal touch. Common edits include correcting technical terminology that Claude got slightly wrong, adding a personal anecdote or context that the AI could not know from the recording alone, reordering sections for better narrative flow, removing coverage of a tangential rabbit hole that is not relevant to the main topic, and adding emphasis markers (like pauses or stress indicators) that guide the voice synthesis.

VidNo's script format supports inline annotations for voice synthesis. You can add [pause] markers for natural breaks, [emphasis] tags for words that should be stressed, and [slower] or [faster] tags to adjust pacing for specific sections. These annotations are stripped from the rendered video's subtitles but influence how the voice model delivers each line.

If you want to skip the interactive review and let VidNo process everything automatically, use the --auto flag. This is useful for batch processing or when you trust the script generation and want fully hands-off operation — VidNo will render all four videos, generate thumbnails, and upload everything to YouTube without any manual intervention.

For users who want even more control, the --script-only flag generates scripts without proceeding to rendering, letting you edit at your leisure before running the render step separately.

Question 18

What screen recording formats does VidNo accept?

Accepted Answer

VidNo accepts virtually any screen recording format you throw at it. The supported container formats are MP4, MKV, MOV, WebM, and AVI. If FFmpeg can decode it, VidNo can process it.

This broad format support means you do not need to change your existing recording workflow. OBS Studio users can keep their default MP4 or MKV output settings. GNOME or KDE screen recorder users get WebM files that work directly. QuickTime exports from colleagues sharing MOV files work fine. Even legacy AVI recordings from older tools are supported.

Resolution support ranges from 720p (1280x720) to 4K (3840x2160). VidNo's OCR accuracy improves with higher resolutions because text is rendered with more pixels. We recommend 1080p as the minimum for reliable code readability, and it is the sweet spot for most developers — high enough for clear OCR, low enough that recording does not impact system performance.

Any framerate is accepted — 24fps, 30fps, 60fps, or variable framerate recordings. VidNo normalizes the framerate during its analysis pass, so inconsistent frame timing from variable framerate recording tools does not cause issues.

One important consideration: VidNo processes the video stream, not just the audio. Tools like Loom, Screencast-O-Matic, or other cloud recording platforms typically let you export or download your recordings as MP4 files. Those exported files work perfectly with VidNo.

Audio within your recording is optional. If your recording includes microphone audio or system sound, VidNo can analyze spoken words for additional context during script generation, but it is not required. The AI narration replaces any original audio in the final output.

For best results, keep these recording tips in mind: use at least a 14px font size in your editor, maintain 1080p or higher resolution, avoid excessive monitor switching (VidNo handles multiple monitors but performs best tracking a single primary display), and close notification popups that might overlay your code content.

File size has no practical limit — VidNo processes recordings incrementally and does not load the entire file into memory at once.

Question 19

How much does VidNo cost?

Accepted Answer

VidNo offers three tiers designed to match different usage levels and needs. Every tier runs locally on your hardware — the pricing difference is about features and convenience, not compute.

The Free tier costs nothing. You download VidNo, provide your own Claude API key (about $0.003 per video in API costs), and get the full core pipeline: screen recording processing, OCR analysis, AI script generation, basic TTS voice synthesis, video rendering, thumbnail generation, YouTube Shorts creation, and automatic YouTube upload with full metadata (title, description, tags, chapters, thumbnail, scheduling). The Free tier includes one voice profile with standard (non-cloned) TTS, 1080p output at 30fps, and all four output formats (tutorial, recap, highlight reel, YouTube Short). This tier is genuinely usable for production content — it is not a crippled demo. The entire pipeline from recording to YouTube publish runs from one command.

The Pro tier costs $29 per month. It adds voice cloning (up to five profiles trained on your actual voice), 4K output at 60fps, priority script generation (Claude Opus instead of Sonnet for higher quality scripts), and custom intro and outro templates. Pro is the right choice when you are publishing regularly and want the voice clone quality bump to make your videos sound like you recorded every voiceover by hand.

The Team tier costs $99 per month and supports up to five team members. Each member gets their own voice profiles and processing queue. Team adds batch processing (queue multiple recordings and process them sequentially overnight), a shared template library for consistent branding, an API for integrating VidNo into your own automation pipelines, and priority support. Team is designed for dev agencies, content teams, and organizations producing developer video content at scale.

All paid plans bill monthly with no annual commitment required. You can upgrade, downgrade, or cancel at any time. Cancellation is immediate — no retention calls, no hidden fees. Your voice models and all generated content remain yours.

The Claude API key is required on all tiers, including paid ones. VidNo does not resell Claude API access — you maintain your own direct relationship with Anthropic. This keeps costs transparent and avoids markup.

Question 20

What can I do on VidNo's free tier?

Accepted Answer

The free tier is a fully functional video production pipeline, not a trial or demo. You can process unlimited recordings and produce unlimited videos at no cost beyond the Claude API charges (roughly $0.003 per video).

Here is exactly what the free tier includes: full screen recording analysis with OCR and git diff detection, AI script generation via Claude (using Claude Sonnet by default), text-to-speech voice synthesis using a standard voice (not your cloned voice), smart editing with intelligent cuts and transitions, four output videos per recording (full tutorial, quick recap, highlight reel, YouTube Short), thumbnail generation for each video, automatic YouTube upload via API with full metadata (title, description, tags, chapters, thumbnail, scheduling), 1080p output at 30fps, script review and editing before rendering, and all supported input formats. The complete pipeline — from screen recording to published on YouTube — runs from one command on every tier.

Here is what the free tier does not include: voice cloning. You get a high-quality standard TTS voice that sounds natural but is not your voice. This is the single biggest difference between free and paid. Many users produce great content on the free tier and only upgrade when they want that personal voice touch.

Other features reserved for paid tiers: 4K output and 60fps (Pro), multiple voice profiles (Pro), batch processing (Team), team member management (Team), API access (Team), custom intro/outro templates (Pro), and priority script generation with Claude Opus (Pro).

To use the free tier, you need to provide your own Claude API key from Anthropic. Sign up at console.anthropic.com and load a minimum of $5 in credits. At $0.003 per video, that $5 deposit processes over 1,600 videos before running out. Claude API billing is purely usage-based with no monthly minimum.

The free tier has no time limit, no watermarks on output, no recording length restrictions, and no artificial processing throttling. Your videos are yours with no VidNo branding unless you choose to add it.

If you are deciding whether to try VidNo, start with the free tier. Process a few recordings, evaluate the output quality, and upgrade to Pro only when voice cloning becomes something you actively want. The free tier is not designed to frustrate you into paying — it is designed to let you make real content.

Question 21

Can I cancel my VidNo subscription anytime?

Accepted Answer

Yes, you can cancel your VidNo subscription at any time with no penalties, no cancellation fees, and no retention process. One click in your account settings and it is done.

When you cancel, here is exactly what happens. Your paid features remain active until the end of your current billing period. If you cancel on day 10 of a monthly cycle, you keep Pro or Team features for the remaining 20 days. VidNo does not prorate refunds for partial months, but you do get full use of what you paid for.

After your billing period ends, your account reverts to the free tier. You keep full access to VidNo's core pipeline — recording processing, script generation, standard TTS, video rendering, thumbnail generation, YouTube Shorts creation, YouTube upload, and all four output formats. You lose access to voice cloning, 4K/60fps output, custom intro/outro templates, and plan-specific features.

Your voice models are exportable before cancellation (and after, while they still exist on your machine). Since voice models are stored locally in ~/.vidno/voices/, they are on your hardware regardless. But the Pro/Team plans include an explicit export function that packages your voice models with metadata for archival or transfer to another machine. Run vidno voice-export --all before cancelling if you want clean backups.

All videos you produced during your subscription are permanently yours. VidNo generates local files on your machine — there is nothing cloud-hosted to revoke. Your content, your hardware, your files.

If you resubscribe later, your existing voice models are immediately re-activated. You do not need to record new voice samples or retrain anything. Your configuration, preferences, and templates are preserved locally.

Team plan cancellation works the same way for each seat. The team admin cancels, and all members revert to free tier access at the end of the billing period.

There is no annual commitment on any plan. Month-to-month billing is the only option. We intentionally chose not to offer discounted annual plans because we want the barrier to cancellation to be zero. VidNo earns your monthly payment by being worth it, not by locking you into a contract.

Question 22

Is my code safe with VidNo?

Accepted Answer

Your code stays on your machine. This is the foundational architectural decision behind VidNo and the primary reason it runs locally instead of in the cloud.

When you process a screen recording through VidNo, here is the complete data flow. Your recording file sits on your local disk and is read by VidNo's local process. Frame extraction happens locally. OCR runs locally on your CPU. Git diff detection runs locally. Voice synthesis runs locally on your GPU. Video compositing and rendering runs locally. The output videos are written to your local disk.

The single exception is script generation. VidNo sends extracted text — not images, not video frames, not audio — to Claude's API for script generation. This text consists of code snippets read by OCR, terminal output text, git diff summaries, and file path information. Think of it as the text someone would see if they were reading your screen, but without any visual context.

Anthropic's API usage policy states that API inputs are not used to train their models and are not stored beyond the processing window. VidNo's request is processed, the response is returned, and the data is discarded on Anthropic's end. You can verify this in Anthropic's current API Terms of Service.

No screenshot or video frame ever leaves your machine. The OCR processing that reads your screen happens entirely locally using Tesseract OCR running on your CPU. VidNo does not use any cloud-based OCR service.

No audio ever leaves your machine. Voice synthesis runs on your local GPU using the locally-stored voice model. There is no cloud TTS call.

Your screen recordings are never uploaded anywhere. VidNo reads them directly from your filesystem.

For developers working on proprietary codebases, client projects under NDA, or anything sensitive, this architecture means VidNo is safe to use on confidential code. The only data that touches the internet is extracted text for script generation, and it is processed under Anthropic's API data handling policies which prohibit training on API inputs.

If even that text transmission is too much for your security requirements, VidNo has a --offline flag that skips Claude and lets you write the script manually.

Question 23

Where is my voice data stored?

Accepted Answer

Your voice data is stored exclusively on your local machine. No voice recordings, voice model weights, or synthesized audio ever leave your computer under any circumstances.

Here is the specific breakdown of where voice-related data lives on your system.

Your original voice sample — the 60-second recording you make during setup — is stored at ~/.vidno/voices/samples/. This is the raw audio file you recorded. VidNo keeps it so you can retrain your voice model if needed without re-recording. You can delete this file after training if you want — the trained model does not depend on the original sample being present.

Your trained voice model is stored at ~/.vidno/voices/models/. This is the MOSS TTS model file that contains the learned weights representing your voice characteristics. The file is typically 200-400MB. This model is what VidNo uses during voice synthesis to generate speech that sounds like you. It runs entirely on your GPU during inference — no network calls, no cloud API, no external service.

Synthesized audio from each video processing run is stored temporarily in ~/.vidno/cache/ during rendering and then in the output directory alongside your final videos. These are standard WAV or AAC files containing the narration for each video. They exist only on your disk.

No part of this pipeline involves any cloud service. MOSS TTS is not a cloud-hosted model with a local API wrapper — it is a fully local model that runs inference directly on your NVIDIA GPU using CUDA. There is no server to phone home to, no telemetry about your voice usage, and no way for VidNo (or anyone else) to access your voice model remotely.

If you use the Team plan, each team member's voice data is stored on their own machine. There is no central server that aggregates voice models. Team functionality is managed through license validation and feature flags, not centralized voice storage.

You can back up your voice models by copying the ~/.vidno/voices/ directory. You can move them between machines. You can delete them permanently at any time by removing those directories. Full control, full ownership, always local.

Question 24

What data does VidNo send to the cloud?

Accepted Answer

VidNo sends exactly one type of data to an external service: text content to Anthropic's Claude API for script generation. Nothing else. Here is a precise accounting of what is and is not transmitted.

What IS sent to Claude's API: extracted text from OCR (code visible in your editor, terminal output, comments, file names as they appear on screen), git diff summaries (text descriptions of what changed between frames, like "+47 lines in src/api/route.ts"), file path information visible on screen, and a prompt instructing Claude to generate a narration script. The total payload is typically 3,000-8,000 tokens of text — roughly 2-6 pages of plain text.

What is NOT sent anywhere: video frames or screenshots from your recording (OCR happens locally, only the resulting text is transmitted), audio from your recording, your voice sample or voice model, the rendered output videos, your local file system contents beyond what is visible on screen, any identifying information about your machine or network, and telemetry or usage analytics of any kind.

VidNo does not have analytics, tracking pixels, crash reporting, or any form of telemetry. There are no background network calls. The only outbound network request VidNo ever makes is the single HTTPS POST to Anthropic's API during the script generation step.

Anthropic's API data handling policy is clear: inputs sent via API are not used to train models and are not retained beyond the processing window. This means the text from your coding session is processed, the script is generated and returned, and Anthropic discards the input. You can review their current data policy at anthropic.com for the latest terms.

If your work involves classified material, highly sensitive client data, or code under strict data handling agreements, you can use VidNo's --offline flag to skip the Claude API call entirely. In offline mode, VidNo generates a structured outline from the OCR analysis and you write (or dictate) the full script yourself. All other features — voice synthesis, editing, rendering — work normally in offline mode.

For the vast majority of developers, the text-only API call represents a minimal and acceptable privacy surface area. No visual or audio content ever leaves your machine.

Question 25

My video isn't rendering -- what should I check?

Accepted Answer

When VidNo fails to render, the cause is almost always one of five common issues. Work through these checks in order — they are listed from most frequent to least frequent.

Check one: GPU drivers and VRAM. Run nvidia-smi in your terminal. If the command fails, your NVIDIA drivers are not installed or not loaded. Install them with sudo apt install nvidia-driver-535 (or the latest version for your card) and reboot. If nvidia-smi works, check the memory usage line. VidNo needs at least 10GB of free VRAM for voice synthesis. Close GPU-intensive applications — games, other AI tools, GPU-accelerated browsers, or CUDA-based programs — and try again.

Check two: disk space. VidNo creates large intermediate files during processing. Run df -h and verify you have at least 50GB free on the drive where ~/vidno/ lives. If space is tight, clear old VidNo cache files with vidno cache-clean, which removes intermediate artifacts from previous processing runs while keeping your final output videos.

Check three: FFmpeg installation. Run ffmpeg -version in your terminal. If FFmpeg is not found, install it with sudo apt install ffmpeg. VidNo requires FFmpeg for frame extraction, audio processing, and final video compositing. A missing or broken FFmpeg installation is the most common cause of rendering failures that occur at the compositing stage (after voice synthesis succeeds).

Check four: Claude API key. Run vidno config --check to verify your API key is valid and has available credits. An expired, revoked, or empty-balance API key causes script generation to fail, which halts the entire pipeline. Log into console.anthropic.com and verify your key is active and has credits.

Check five: recording file integrity. Run ffprobe your-recording.mp4 and verify it reports valid video stream information. Corrupted or truncated recordings — often from OBS crashes or abrupt recording stops — can cause VidNo's frame extraction to fail. Try processing a different recording to isolate whether the issue is with VidNo's installation or the specific file.

If none of these checks resolve the issue, run vidno process your-recording.mp4 --verbose to get detailed pipeline logs. The verbose output shows exactly which stage failed and the specific error message. Common verbose-mode findings include CUDA out of memory (reduce other GPU usage), permission denied on cache directory (fix with chmod), and unsupported codec in recording file (re-export from OBS as MP4/H.264).

Frequently Asked Questions

General

Setup & Requirements

Voice & Audio

Video Output

Pricing & Plans

Privacy & Security

Troubleshooting