Flowvdo: From a Story Idea to a Finished Google Flow Video

The first clip you make in Google Flow is exciting. The tenth is work.

Say you've written a little story about a fox who delivers letters through a snowy village. Scene one is lovely. Scene two has a fox with a slightly different face. By scene four the scarf has changed colour, the village has turned into a city, and you've lost track of which prompt produced which file in your downloads folder. Then you line up eight clips in an editor and hope the lighting matches.

Google Flow is built around the single generation. A story needs a plan, a cast that holds still, a lot of patient clicking and a final file. That gap is what we built Flowvdo to close.

What Flowvdo actually does

Flowvdo is a Chrome extension with a companion web app. You give it an idea, and it does three jobs:

1. It plans the story with an AI provider you connect: outline, characters, places, every scene and a Flow prompt for each one.
2. It runs every scene in Google Flow, in your own Google account, filling in prompts and references and saving each clip as it finishes.
3. It joins the clips in order into one MP4, on your computer.

One distinction matters before anything else. Flowvdo doesn't generate video. Google Flow does, with your Flow credits. Flowvdo is the planner and the operator sitting beside it, and it never buys, sells or meters those credits. Your AI key, likewise, is yours: you pay your provider directly. Flowvdo isn't affiliated with Google, and Google's terms still apply to your account.

Planning before spending

Most wasted Flow credits come from improvising: generate, notice the story doesn't work, rewrite, generate again. Flowvdo moves that thinking to the front, where it's cheap.

You start with a brief. The idea itself needs only a sentence, but you can also set the audience, tone, genre, visual style, language, target length, aspect ratio (16:9 or 9:16) and which Flow model, resolution and clip length to use. There are also six templates: a product ad, an explainer, a bedtime story, a vertical social short, a music video and a travel postcard.

Then your AI provider works through four separate steps. The first is an outline with a title, logline, a beginning, middle and end, the characters, the places and a "visual bible" for style, palette and lighting. The second breaks that into scenes, each with an objective, action, camera, place, cast and duration. The third writes a Google Flow prompt for every scene. The last is a continuity read-through that lists problems, like an outfit that contradicts a character's locked traits or a place that's missing a reference.

None of this touches Flow. You can press Refine with AI for a shorter, longer, funnier or more dramatic version, or type your own instruction. You can edit any scene by hand, down to the exact prompt, or regenerate just one scene. Every change becomes a plan version. When it reads right, you approve it, and that version is locked for the run.

Keeping the fox the same fox

Character drift is the complaint we hear most about multi-scene AI video, and there's no single trick that fixes it. Flowvdo stacks three.

At the start of a run, it generates a reference sheet for each character (front, side and back) and an establishing shot for each place, right inside your Flow project. Those images are attached to every scene the character or place appears in. They're generated under a zero-credit cap, so if Flow ever shows a cost for them, the run stops and asks you first.

Next, locked traits. Each character has a short list of phrases that must never change, such as "red scarf" or "orange antenna light", and the planner repeats them in every video prompt. The visual bible adds a style line to the start of every prompt too, so the look carries from scene to scene.

Finally, last-frame chaining. When you start a run you choose how scenes connect: each scene fresh from its references, the previous clip's last frame added as an extra reference, or the next scene starting directly from that last frame. The frame is extracted on your machine, and chaining costs nothing extra.

It isn't magic, and Flow still makes the final call on every pixel. But it's the difference between hoping for consistency and setting it up.

Watching a run

Before anything is spent, a readiness checklist confirms the extension is connected, your account is fine, and there's a signed-in Flow tab. Flowvdo also checks that it can find every control it needs on Flow's page. If Google has changed the layout, it stops before spending a credit rather than clicking blindly.

Then it opens its own Flow tab and works through the plan one generation at a time, at least five seconds apart. A small Flowvdo button appears next to the project title in Flow, with a floating tracker showing characters, places, scenes and the final video, and the same live progress shows in the side panel and the web app.

You choose how generation is submitted. In Automatic mode, Flowvdo presses Generate with one real click. In Manual mode it fills in the prompt, settings and references, then waits for you to press the button yourself.

The part we're most careful about is cost. You see an estimate before you start. Before every generation, Flowvdo reads the credit cost Flow actually shows, and each run has a budget, by default the estimate plus 10%. If the next step would go over, the run pauses.

It pauses for other reasons too, and it always says which. Flow wants you to verify you're a person? You complete that yourself; Flowvdo never interacts with Google's checks. Signed out, out of credits, or at the model's daily limit? The run waits with a one-line explanation of what to do. If Chrome restarts mid-run, the extension picks up where it left off and looks for a scene that was already submitted before generating anything again, so you don't pay twice for the same clip.

One video at the end

Each clip is saved and checked the moment its scene finishes. When the run is done, Flowvdo joins the clips in story order inside the extension. If they share a format, which is the normal case, they're joined without re-encoding. If a clip is missing, it's left out and listed, rather than holding up the whole video.

Everything lands in a folder you choose (or Downloads/Flowvdo/<project>/ if you don't): character and place references, every numbered clip, the final MP4, and a manifest.json with each file's size and SHA-256 checksum. Nothing is uploaded to Flowvdo.

Not just stories

Plenty of Flow work isn't a narrative at all, so Flowvdo has two other modes running on the same engine, with the same budgets and pauses.

Batches are lists of up to 300 prompts, images or videos, one to four outputs each. Paste them, import a .txt file, or give your AI provider an idea and a number and let it write them. References can go to every prompt, to each in turn, or only to the ones a prompt names with an @mention like @mug. File names follow a template you set, and Run again can redo only the failures.

The Workflow Builder is a node editor for when you want a repeatable pipeline: inputs, an AI Agent node, Flow Generate, Last frame, Angles, Effects, logic nodes and Assemble video. It comes with five templates, including one that turns a product photo into four angles and then a video of each, and one that goes from a character sheet to three chained scenes to a final video.

Supporting both is a library: reference pictures with @slugs that stay in your browser until a prompt needs them, saved prompts with {variables}, styles appended to every prompt, and tools that turn one picture into several camera angles or restyle it as watercolour, claymation and so on.

Bring your own AI

Planning, prompt writing and the AI Agent node all run on a provider you connect in the extension's settings. The list includes OpenAI, Anthropic Claude, Google Gemini and OpenRouter, plus any OpenAI-compatible endpoint, with presets for Groq, xAI, DeepSeek, Mistral and Together AI, and local models in Ollama or LM Studio.

Revoye, our own OpenAI-compatible AI API, sits at the top of that list, and it's a good fit for planning. The one thing it can't do is describe images, so for the "describe this picture" features you'll want a vision-capable model from OpenAI, Anthropic, Gemini or OpenRouter instead. If you're curious how Revoye works more broadly, we wrote about adding AI to an app with one API.

Your key is encrypted in the extension with AES-256-GCM and sent only to your provider. Flowvdo's servers never see it.

For people who live in AI assistants, Flowvdo also has an MCP server. With a personal access token scoped to exactly what you tick, Claude or another MCP client can list your projects, create batches, start runs and check on them. The run itself still happens in your paired extension, with your budget.

Who we made it for

We had a few people in mind: YouTube creators running story or faceless channels who need a recurring cast, marketers making product ads in Flow who want twenty variations rather than one, and AI filmmakers who want a pipeline they can rerun. And honestly, anyone who has made a great single clip in Flow and wondered how people turn those into actual videos.

What it won't do

Flowvdo stops at a finished file. It doesn't add music or voice-over, and it doesn't publish anywhere. It also saves the files at the size Flow makes them; it doesn't upscale. You need Chrome or another Chromium browser on a computer, and a Google account with Flow access.

For the publishing half, our own Vdoup takes one video to YouTube, TikTok, Instagram and the rest with captions written for each platform, so the two fit together neatly: Flowvdo makes the MP4, Vdoup gets it out.

Plans

Flowvdo's plans limit projects, runs, scenes and video length. Your Flow credits and AI usage are separate and stay with Google and your provider.

The Free plan gives you one active project, two runs a month, and videos of up to four scenes and 30 seconds. That's enough to make a real short and see the whole thing work. Creator goes to ten projects, 30 runs a month and videos of up to 20 scenes and three minutes. Pro removes the project limit and allows 150 runs a month, up to 60 scenes and ten minutes per video, with priority support. Stories, batches and workflows share the same limits, and each batch item counts as a scene. Current prices for the paid plans are on the Flowvdo pricing page.

Try it on your next idea

Pick a story you've been meaning to make, even a thirty-second one, and run it end to end on the Free plan. Seeing an approved plan become one finished file is the quickest way to tell whether Flowvdo belongs in your workflow.

You'll find the details and setup guides at flowvdo.cc, and a full overview of the features on our Flowvdo product page.