Google Flow Automator StudioUser Guide

Project Title and Project Summary — what should I write?

Project Title is just the name used to identify the project in the project list.

Project Summary is the single most important field for good results. Describe what you want to create: the overall scene, the visual style, the tone, and anything else that helps the AI understand your concept. The more context you give here, the better every generated result will be.

Story Style — when should I change it from Auto?

Leave it on Auto if your Project Summary is detailed enough. If your summary is thin, pick the style that best matches the video you want and you will get better results.

Frame rate and Video Format (16:9 / 9:16)

Frame rate is used for export. Leave it at 30 fps in most cases.

Video Format is 16:9 for long-form and 9:16 for short-form. Studio selects this automatically from the media you import. Once a project has content the format is locked in Project settings — change it from the format control in the toolbar instead, which asks you to confirm before refitting every scene.

Export Prompt CSV / Import Prompt CSV — filling in missing prompts

Studio uses the image / video prompt as its primary source for understanding your media. If your GPU is powerful enough to run Local AI, the AI will also analyse the images themselves on top of the prompts.

If some of your media has no prompt attached, or the prompts went missing during import, export the CSV, fill in the missing prompts by hand, and import it back. Much faster than editing scenes one at a time.

The Project menu — settings, CSV, export, save, project list

Everything that applies to the whole project lives under Project in the toolbar: Project settings, Export / Import prompt CSV, Export, Save and Project List.

Studio saves automatically after every step, so you do not need to save manually — though clicking Save before you close is good practice. Project List returns you to all your projects; your current work is already saved.

How do I export the finished video?

When your edits are done, click Export. You can render an MP4 ready to upload, or export all the individual assets to carry on in an external editor. See Exporting for the full list.

What do I have to fill in?

• Project Title — the name used to identify the project in the project list.
• Project Description — the same field as Project Summary in Project Settings. A short summary of your project and the visual style you want.
• Language — the language used for captions and narration.
• Story style, AI writer, Image analysis, Voice and Background music — all have sensible defaults, so you can press Generate straight away.

Which AI engine should I choose?

• Local AI — the local model you installed, in three sizes. Free, but limited in reasoning.
• OpenAI / ChatGPT — your own API key.
• Claude — your own API key.
• Gemini — your own API key.

All API keys are stored locally. If your PC cannot run Local AI and you do not want to pay for an API, use Copy Prompt for ChatGPT/Claude/Gemini instead — it is free.

Which voice option should I choose?

• Local TTS — two local providers: Local TTS and Local TTS Asian (Korean and Japanese). Together they cover 10 languages. Free, but the choice of voices and tones is limited.
• ElevenLabs — your own API key, for many more voices and languages.
• OpenAI TTS — your own API key, for more voices and languages.
• No narration audio — if you plan to record your own voice afterwards.

How do I control the background music?

• Automatic — the local BGM service picks a style and generates it for you.
• Manual — describe the BGM style you want.
• No background music — if you do not want music, or want to upload your own later.

What is in the AI menu?

Four tools for the project you already have — Enhance, Narration / Caption Editor, Generate Approved Narrations and CC extraction from video — plus the free chat-AI route underneath.

Enhance — improving a result you don't like

Describe what you want changed and regenerate. Use it when the output is close but not right, rather than starting the whole project again.

Narration / Caption Editor — editing everything at once

Shows every narration and caption in the project in one place, so you can review and rewrite them together. Make your changes, finish editing, then continue to the next step to regenerate TTS from the new text. Captions can also be exported and imported here.

Generate Approved Narrations — regenerating every scene

Regenerates narration for all scenes at once.

To update narration by hand, edit it in the Audio tab of each scene. You can then regenerate that single scene's audio there, or use Generate Approved Narrations to redo every scene in one pass.

CC extraction from video — captions from existing audio

If all your scenes already have narration, this extracts captions from the existing audio using the local captioning model.

Copy Prompt for ChatGPT/Claude/Gemini — free, no API key

Local AI is limited in reasoning compared with cloud models. If your PC cannot run the Medium local model, this lets you borrow a cloud LLM through its ordinary chat interface — no API key and no extra cost.

Choosing it opens a dialog with three buttons. Press them in order.

1. Copy Project prompt assets. The complete brief goes to your clipboard. Paste it into ChatGPT, Claude or Gemini exactly as it is — add nothing of your own.
2. Open Snapshot. Drag every image in the folder that opens into the same chat message, before you send it. The pictures are what let the model see your scenes, so the answer comes back better.
3. Send it and wait for the full reply. Then copy all of it, including the code block at the end, and press Paste Cloud LLM response.

From here it works exactly like One Click generation.

AI Server On / Off — why isn't it running?

The AI server is not running when Studio starts, because it is only needed when the AI has to analyse images. Turn it on or off as required — leaving it off saves memory while you are only editing.

Feedback & Log — sending a problem report

Send feedback, or report a problem with a description of what went wrong. Please tick Attach privacy-safe diagnostic logs — these contain only information about how Studio is running. No API keys and no media are ever transmitted.

Scene Name and Duration

Name identifies the scene. Duration is its length, and it varies with the story and the narration — adjust it in the timeline if you need it longer or shorter.

Image prompt — the field that decides your results

This is the prompt that produced the image or video, filled in automatically during import. If it is missing, the AI cannot identify what is in the shot.

With a strong PC, Local AI will try to analyse the image instead, but local vision is limited and much less accurate. So if a prompt is missing, enter it manually. Whenever results are not what you expected, check the image prompts first. To fill in a lot of them at once, use Export / Import Prompt CSV.

Reanalyze — having the AI look at the image again

Runs image analysis again. Quality depends on the local model you are running; a small model may not be satisfactory. Always enter the image prompt before running Reanalyze.

Source-to-target framing — repositioning and scaling

Repositions and scales your image or video inside the frame. The background option is useful when your asset does not fill the frame — for example a 16:9 clip inside a 9:16 video.

Photo Motion — movement on still images

Applied automatically during One Click generation, and you can also apply it by hand. Photo Motion works on images only, not on video.

Audio — script and voice service

Script is the narration text the TTS engine reads out, generated automatically during One Click generation. Edit it here for a single scene, or in bulk through Narration / Caption Editor. After changing the script, remember to regenerate the voice.

Voice service offers Local TTS, OpenAI, ElevenLabs, recording your own voice, uploading an audio file, or no audio. Local TTS is free but limited; for more voices and languages, add your API key in Settings. Apply voice to all pushes the chosen voice to every scene, and Lock audio protects a take you are happy with.

Subtitles — and fixing timing that has drifted

Subtitles are split automatically, so you normally do not need to touch them. But if you changed the audio, the timing can drift out of sync. Drag and drop them in the timeline, or use CC Retiming in the Audio tab.

Subtitle Style — and applying it to every scene

Change the style and apply it to the current scene, or use Apply style to all scenes to update the whole project at once.

Overlays — text, stickers, emoji, images, templates

Five overlay types are available: Text, Sticker, Emoji, Image and Template. Browse to the one you want, hover over a style, and click Add to place it on the scene.

Text content is edited in the Overlay panel; the position of any overlay is adjusted directly in the preview window.

BGM for a single scene

Background music is generated automatically during One Click generation, or you can generate it manually per scene. To change the mood, generate a new style or upload your own BGM file to that scene.

What can I edit in the timeline — and what can't I?

The timeline is deliberately simple. Studio is not a video editing program, so it does not support splitting clips or fine-grained editing.

• You can change the duration and position of the scene visual, subtitles and overlays.
• You cannot change motion, narration or BGM here — those are controlled in their own tabs.

If you need frame-accurate editing, export the assets and finish in CapCut, Premiere or DaVinci Resolve.

Language, Membership and User ID

• Language — the language of the Studio interface.
• Membership — Professional only during the beta.
• User ID — the same User ID as in Google Flow Automator.

Project AI Models and Managed Local AI (Small / Medium / High)

Project AI Models is the preset that records which AI model this project uses.

Managed Local AI is the local model size. Choose Small, Medium or High to match your computer. Running a model larger than your hardware can handle will make generation extremely slow, or fail outright — not sure which to pick? Run the System Check.

Low VRAM Mode — what it does and when to use it

Local AI needs serious computing power, especially GPU. If your machine is below the minimum requirement, this mode turns on automatically. With it on, Local AI is skipped and you use a cloud LLM such as OpenAI or Claude instead — including the free Copy-for-LLM route that needs no API key.

This is not a downgrade for most people. Local AI is limited in both power and reasoning, so cloud models often give better results anyway. You can also turn Low VRAM Mode on manually even with a good GPU.

The other local models — captioning, BGM and TTS engines

• Local Captioning Model — used when audio has to be analysed to extract captions. Not needed unless your video has narration, but it will be used more in future updates.
• Local BGM — generates background music. It also needs significant computing power, so generation time depends on your hardware.
• Local TTS Engines — generate speech locally. Each supports a different set of languages, so pick the one you need, or install both for maximum flexibility. Voices and tones are limited; connect an API key for a cloud service such as ElevenLabs if you need more.

API keys — do I need them?

No. Local AI, TTS and BGM are limited but free to use inside GFA Studio. For more flexibility, add your own API keys to use cloud AI and TTS — but that spends your own credits with those providers. All keys are stored locally on your machine and are never sent to a GFA server.

Advanced — Activity log and usage sharing

Activity log is off by default. Turn it on if you are having trouble with Studio; we will ask you for the results when diagnosing the issue.

Share privacy-limited Studio usage counts shares basic usage counts so we can see how Studio is used and improve usability and performance.

Restart Studio and Uninstall Studio

Restart Studio if you run into a problem. Closing and reopening does the same thing, but shutting down the AI server can take a while — restarting from here reboots the server too.

Uninstall Studio when you no longer need it and are ending your membership. Your project folder and exported files stay on your computer.

What exactly can I export?

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.

Can I take the export into CapCut, Premiere or DaVinci Resolve?

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.

How is the correct order guaranteed with hundreds of files?

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.

If I stop using Studio, do I lose my work?

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.

The results aren't good — where do I start?

1. Check the image prompts. A scene with no prompt is invisible to the AI. By far the most common cause.
2. Rewrite the project description. A thin description produces a thin story.
3. Use Enhance to fix a specific result rather than regenerating everything.
4. Try a stronger engine. Local AI reasons less well than a cloud model — the free Copy-for-LLM route often lifts quality noticeably.

Generation is extremely slow, or it fails partway

Almost always a model that is too large for the machine. Lower Managed Local AI one step, or switch on Low VRAM Mode and use a cloud LLM. Close other heavy applications while Studio is generating — they compete for the same GPU and RAM. Run the System Check to see what your PC is suited to.

The subtitles are out of sync with the narration

This happens after you change the audio. Drag the subtitles in the timeline, or use the CC retiming feature to line them up again.

I edited the narration but nothing changed

Changing the script does not regenerate the audio by itself. Regenerate the voice for that scene in the Audio tab, or use Generate Approved Narrations in the AI menu to redo every scene.

Studio won't open from the extension

Confirm the extension is signed in with the account that holds your membership — the connection is established from the extension side. Remember that Trial and Starter accounts cannot open Studio.

Studio has become unresponsive

Use Restart Studio in Settings. It reboots the AI server as well, which a plain close-and-reopen may leave hanging.

How do I report a problem so it can actually be fixed?

Use Feedback & Log in the AI menu. Describe what you were doing when it went wrong and tick Attach privacy-safe diagnostic logs. If the problem is reproducible, turn on the Activity log in Settings first, reproduce it, then send the report.

How rough is the beta?

It is a beta and we are not going to pretend otherwise. The 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.