Google's 3 Free Video Tools Actually Work? Vids, Flow and Pomelli Tested

Let's be clear up front: I ran more than ten free Google video tools through the same briefs, five produced nothing usable, two were weak, three actually delivered. In this video I walk through those three — Google Vids, Google Flow and Pomelli — with practical setup so you can run everything in the browser, no installs and no cost, but each has a different superpower and a different ceiling.

Vids.new: The Fastest Start In The Browser

The fastest start is always Google Vids because unlike open-source video models you install nothing; you type vids.new and land in a new project. It first asks for orientation — landscape, portrait or square — I picked landscape for maximum flexibility. The grid offers record yourself, upload files and turn slides to video, but the three strongest controls live in the right panel: create AI videos, AI avatar and editing. For a quick test I skipped reference images, pasted a plain prompt and got a surprisingly faithful result.

I asked for a man running toward the camera with pigeons bursting up; motion had a slight delay yet framing and tracking were correct and for a free tool it beats most rivals. This isn't just my take — OpenArt recently launched OpenArt Arena, a blind leaderboard where pros rank every major model, and there Google's Omni Flash sits in the top five for video while Nano Banana Pro sits in the top three for images. For models you can use free, that's striking and a useful compass before you commit.

Edit Or Recreate?

Under each clip Vids shows three actions: extend, edit and recreate. Recreate makes a brand-new clip, edit keeps your clip and changes only the detail you describe — most people break a good clip by hitting recreate when they meant edit. I used edit to turn the shirt red and switch to night; everything else stayed coherent and I fixed it with one line instead of opening an editor. If you like the clip and need one detail changed, edit is the right button.

The third strong piece is the AI avatar: I hovered the voice presets, picked Eric, pasted a script, hit preview and then generate. I had him read a short piece about how film cost used to be stock and processing, why nobody shot ten takes unless needed, and how digital killed that cost so the expensive part became knowing what to shoot. Avatars support per-scene or full-project generation, so from automated explainers to short verticals Vids alone works like an end-to-end free studio — yet advanced cinematic control lives in Flow.

Flow's Secret Weapon: The Agent

What makes Google Flow special is the agent inside; you don't need to craft a perfect prompt because the agent plans for you. I opened a new project, switched the prompt box to agent and expanded it into a full side panel — you can still go manual via the model name if you want. I set two one-time defaults: confirm before generating to always, image to 16:9 with Nano Banana 2 and video to 16:9 with Omni Flash, then asked the agent in plain English for a wide cinematic image with Nano Banana Pro. It narrated its plan, showed thinking when expanded, and delivered a realistic frame of a woman appearing to repair a plane after landing in a salt desert — high fidelity because the agent handled the details.

On video Flow offers two models and your choice directly shapes the output. Veo 3.1 quality is the top tier but caps at eight seconds and has no separate setting for shot cuts; you describe cuts as timed beats inside the prompt, which Google calls timestamp prompting. For the fish market scene I wrote each shot with its time range and action, added cut cues between them and closed with what must stay constant — market, light and characters — and the agent paused for approval, summarizing the job; that's why I keep confirm always on. The resulting clip had impressive cinematic light and cutting, though realism, especially the man at the start, still felt off.

Eight Seconds Or Ten? And Keeping A Face Consistent

For the second model I let the agent decide: I asked for a ten-second hand-drawn storybook clip of a small orange fox, the agent caught that Veo 3.1 tops at eight seconds and offered two paths — switch to Omni Flash or shorten — I switched and got a consistent ten-second animation without trimming details. Lesson is simple: need longer, whimsical duration, pick Omni Flash; want peak cinematic punch in short bursts, pick Veo 3.1. For longer stories the real blocker is face consistency and Flow solves it with a character element; I built Halver from scratch via a text description, got a stern northern European type, gave him the Umbreel voice and a personality note, saved him, then invoked him in the agent chat with @ and directed a ten-second Omni Flash scene with a line — forty years on that water, it never owed me anything — the delivery carried the personality note and the face stayed locked.

Pomelli: From One URL To Brand DNA

The surprise of the set is Pomelli; it doesn't start from a prompt but from a website and analyzes the brand so every output looks native to it. I fed it Oatley, whose style I like, though you can also build a DNA from up to twenty uploaded files without a site; after a few minutes it returned a business DNA of typeface, palette, values and voice, all editable. From there I generated multiple site drafts, an auto-built brand book that pulls real product photos from the site after you pick a cover and confirm up to seven images, and a photo shoot; for the O drink I picked six templates at story 9:6 and hit generate, getting six premium variations, each animatable to video. Finally I typed a one-line campaign brief and it returned an editable title, description, goal and offer for refinement before rendering a package whose copy was witty and whose visuals stayed on brand — strikingly sellable for a free tier.