From Flow clips to a
finished video.

The rendered MP4; every scene clip as a separate file; narration audio per scene and as one continuous track; the background music track; captions as standard SRT; the original prompt for every scene as a CSV; and a manifest describing scene order, start times and durations.

Yes. Everything exports as standard media files plus SRT captions, which every major editor imports. The manifest and zero-padded file names keep hundreds of clips in the right order wherever you drop them.

File names are zero padded — scene_001, not scene_1 — so alphabetical sorting matches scene order in every tool. The manifest carries the exact timing if you need it.

No, and this is deliberate. Projects live in a local folder you choose, and the export is ordinary media files rather than a proprietary project format. There is nothing to trap you here.

Studio analyzes each image together with its original prompt to understand the full context and content. It's reading not just what's in the picture, but matching it against what you asked for, so everything makes sense when the narration is written. The more detailed your prompts and the more complex your images, the longer this step takes.

The "one click" creates a finished video by running several steps inside: image analysis (reading your prompts and media), story generation (writing a script that connects all your scenes), narration generation (turning that script into audio), caption generation (timing captions to match the audio), and background music generation. Each step is happening on your computer, and all of it is heavily influenced by your computer's processor, RAM, and GPU. A faster machine will finish noticeably sooner. This is not a network delay — your hardware is doing the work.

Performance scales with your CPU, GPU, and available RAM. Upgrading to a newer processor, adding more RAM, or using a computer with a dedicated graphics card can cut generation time significantly. The AI models are the bottleneck, not the network or our servers.

You have two options. First, you can regenerate it — select a scene, change the narration text if you want, and click regenerate. If you're using a local AI model and it's not meeting your needs, you can switch to cloud narration: use your own OpenAI, Anthropic, or ElevenLabs API key and Studio will send the prompt directly to those services instead. Your keys stay in your OS credential store and are never stored by GFA.

Yes, completely. The Studio audio editor lets you click to seek, drag to select, trim from either edge, delete a section, and undo. The original recording is never modified — all edits are non-destructive. You can also delete the narration track entirely and record your own voice directly into Studio.

Local text-to-speech models are smaller than cloud services to keep file sizes reasonable on your machine. If you need more voices or higher quality narration, use a cloud provider: connect your own OpenAI, Anthropic, or ElevenLabs key through Studio. That will draw on their much larger model and voice collections. Be aware that using cloud services consumes your API tokens — check the pricing on those services to understand the cost per character narrated.

You can edit the narration text directly in the Studio editor. Highlight any scene, change the text, and regenerate just that scene's narration. If you want to rewrite the whole script, you can also copy the entire script to an LLM like Claude, ask it to rewrite it, then paste the new version back into Studio and regenerate all narration at once. That approach gives you full creative control while keeping the videos and timing in sync.

No. The local AI models work well for most cases and cost nothing beyond the initial download. Many users get excellent results without ever touching API keys. Try local first — if the quality meets your standard, you're done. Use cloud services only if you need more, not by default.

A Pro membership on the Google Flow Automator extension, signed in with the same account. Studio launches from the extension.

If you join while the beta is open, Studio stays included with your Pro membership for as long as you keep that membership — including after Studio becomes a separately paid product.

We haven't set the price yet, and we'd rather decide it after seeing how people actually use the beta than announce a number we might have to change. What we can commit to: if you join during the beta and keep your membership, it costs you nothing extra.

Your Studio access ends with your membership. If you rejoin later, whatever terms apply at that time will apply to you — the beta benefit isn't held open indefinitely.

No. A failed payment isn't treated as leaving. While the payment is being retried your access continues, and there's an additional grace period after the billing period ends. The benefit is tied to cancelling, not to a bank hiccup.

Your membership pays for the extension; Studio is a benefit included during the beta rather than a separate purchase, so we don't offer refunds on the basis of Studio. You can cancel your membership at any time, and if something has genuinely gone wrong, talk to us on the support page — we'd rather sort it out than argue about it.

We don't have a fixed date yet. We'll announce it at least 30 days in advance so nobody is caught out.

Running an AI model locally is genuinely demanding work, and the part that decides whether it is pleasant or painful is GPU memory (VRAM). The model has to fit in it. When it does, your graphics card does the work and things move quickly. When it does not, the work spills onto the CPU and system RAM, and the same job can take many times longer.

This is why the graphics row deserves more attention than the others. A fast CPU does not compensate for a card that is short on VRAM, and integrated graphics (Intel UHD / Iris Xe, AMD Radeon Graphics) will not run local AI at all — it needs a dedicated GPU with 6 GB of VRAM at minimum, 8 GB to be comfortable.

You are not shut out. Turn on Low VRAM mode, which stops Studio from loading a heavy local model and lets you use ChatGPT or Claude through their ordinary chat interface instead — you copy the script across rather than running a model on your own hardware.

There is no API key and no extra cost involved in that route. It uses the same free chat you would use in a browser, so a modest PC can still produce the whole video.

Yes. If you already have an OpenAI, Anthropic or ElevenLabs key you can connect it and Studio will call those services directly, which is faster and needs nothing from your GPU. Keys are kept in your operating system's credential store and never sent to a GFA server.

Be aware that this spends your own tokens. Every narration or script request is billed to your account by that provider, so check their pricing before you run a large project. The local and Low VRAM routes remain free — an API key is a convenience, never a requirement.

Yes. The Windows installer is securely signed by Microsoft, which means Windows can confirm it came from us and has not been altered since it was built. You will see the publisher named in the install prompt rather than an "unknown publisher" warning.

Download it only from this page. Copies from third-party download sites are not covered by that guarantee.

No. This release is Windows 10 and Windows 11 only, and there is no macOS build planned for it. Please don't subscribe for Studio if you're on a Mac.

12 GB with local AI, and 20 GB on an SSD is the comfortable figure — most of that is the local model files. If you skip local AI and use cloud AI or Copy-for-LLM instead, 2 GB is enough.

To generate the media in the first place, yes — Studio works with what Flow produces, and Flow's own limits apply to generation. Studio itself doesn't consume Flow credits.

Not to us. Selected media is copied into your local project folder and rendering happens on your machine. Narration is written from your saved prompts, so the images themselves aren't sent to an AI provider to write the script either.

In your operating system's credential store. They are never sent to a GFA server. They travel over TLS only to the provider you chose, when you use that provider.

Yes. Local narration, local captioning and local background music run entirely on your computer once installed. Cloud voices need a connection, and the extension checks your membership periodically.

It's a beta and we're not going to pretend otherwise. Core workflow is solid; some rendering and import edge cases are still being fixed, and details will change between builds. Keep backups of anything you care about and check output before publishing.

Yes, please do. Studio has a feedback and feature request form built in — open Help inside the application and submit it from there. Tell us what you actually want, in your own words.

Every request gets read. If it is something that can be built, it gets added — beta requests are what shape the release version.