Automation — MCP for Google Flow, Google Vids & Claude: AI video autopilot

OmniFlow is a Chrome extension for batch video generation in Google Flow (Veo, Omni 1.1 Flash) and Google Vids. You describe the scenes in your own language, the extension writes the prompts itself, runs the generations one after another automatically and downloads the finished clips. In Google Vids it also makes talking AI presenters from your script. Everything runs on your own Google plan, with no API keys and no per-second billing. Anyone who has used Flow seriously knows the problem: one prompt, one eight-second clip, and again from the start. Twenty clips for a video turn into an evening at the keyboard, and half of the credits go to prompts the model understood differently. OmniFlow takes that routine away: you decide what should be in the frame, the extension does the rest. WHAT THE EXTENSION DOES Saves money. Generations go through your Google AI Pro or Ultra plan, using the same credits you spend by clicking. An eight-second clip through the Veo API costs from 1.2 to 6 dollars; on the Pro plan the same clip comes to about 40 cents. That is 3–8 times cheaper, and even more on Ultra. Saves time. Describe the scenes, press Start and go do something else. The extension starts each generation itself, waits for the result, collects the clip and saves it to a folder with the scene number and name. You come back to a finished batch, not to a tab. Adapts to your task. Write what you need: "headphones ad for a marketplace, 20 seconds, vertical". The extension drafts the script, splits it into scenes, writes a clear prompt for each one by the Omni 1.1 Flash rules (action, shot and camera move, location, style, light, sound) and picks the moves. Write in your language, English goes to Flow. Lines, captions and character names are carried over word for word. 39 cinematic moves at one click. Camera: orbit 360°, dolly zoom, frozen time, FPV flight, crane up. Product: hero rotation, levitation, liquid splash, turntable. Transitions: whip pan, match cut, before and after, logo finale. Images: white background for a marketplace card, remove clutter, restore an old photo, collectible figure. One click, and the exact wording the model responds to is in the prompt. The moves can also be inserted straight into the Flow field if you shoot by hand. Keyframe transitions. Set the first and the last frame of a scene and Omni 1.1 Flash builds the motion between them: day turns into night, a product assembles into a logo. Scene stitching. The next scene starts from the last frame of the previous one, so the hero and the place stay the same from clip to clip. Separate generations become one continuous video. References. A photo of a product or a person for all scenes or for one, characters from the Flow library by @name, a video reference to take the motion or the style from. One photo is enough for Flow to reproduce exactly this item. Sound. Three layers: the clip's own sound, music, voice-over. If the hero speaks in the frame, Flow itself says the line and the lips match. Download. Clips arrive in a folder in the right order. If something did not download, the "grab finished clips from Flow" button fetches them from the project without re-shooting. GOOGLE VIDS: A SECOND SOURCE OF VIDEO Google Vids has its own monthly allowance of seconds that does not touch your Flow credits. When the Flow credits run out, shooting goes on in Vids with the same Omni model. It covers clips from text and from photos (up to three ingredient images), bringing a picture to life, editing a finished clip by description, extending it and upscaling it to 1080p. Any length from 3 to 10 seconds, 9:16 or 16:9. Talking presenter. Vids makes an AI avatar that reads your script with matching lips in 30 languages. Pick the presenter by name (Finley, Sebastian, Mia and others), write the text, and in one to three minutes you have a 1280×720 video. The video is as long as the script. Avatars do not spend clip seconds. It works through Claude: you write "film a presenter who talks about the new collection", Claude prepares the text, picks the avatar and saves the video to your folder. Keep the Chrome window with the Vids tab on screen while the avatar renders. THREE WAYS TO WORK 1. Studio. Manual mode: a map of eight actions (scenario, prompt, camera, effects, avatar, stitch, sound, frames). Each node shows its state in words, the highlight tells you what to fill in next. 2. Assistant. A chat inside the panel on your own Anthropic key. You write "I need a coffee shop ad, 30 seconds, vertical", the assistant clarifies a couple of details, splits the video into scenes, writes the prompts and starts shooting. 3. Claude over MCP. One button in the settings copies an instruction for Claude Code or Claude Desktop. Claude starts a local bridge and works on its own from there: "make a forty-second hotel promo from the photos in the folder, vertical, logo at the end". Claude decides where to shoot, in Flow or in Google Vids, and can add a talking presenter. SAFETY AND YOUR GOOGLE ACCOUNT The extension presses the same buttons you would, with pauses of 25–70 seconds between generations and a twenty-minute break if Flow asks to wait. It has no access to passwords or payment details, you simply stay signed in to your Google account. It works only on the Google Flow and Google Vids pages and on lingoflow.pro. The Anthropic key for the assistant is stored in your browser and is not sent anywhere else. WHO IT IS FOR People who have just opened Google Flow and want a good-looking video on the first try: you write the idea in your language, the extension makes the script and the prompts, step-by-step tips guide you through the task (product ad, brand video, story). People who shoot in series: ads, marketplace cards, content for clients. Dozens of scenes in an evening, editing of finished videos, control through Claude. PRICE The first 24 hours after install are free, no key and no card. Then a key from the website: 29 euros a year …