Google Flow Music Review: Songs, Videos and Lip Sync

Google Flow Music generated original songs and downloadable 720p videos in our tests. The later videos kept a recognizable singer and setting, but the mouth movements did not consistently match the vocals on playback. We also encountered credit rejections despite estimates below the displayed balance.

For song ideas and short visual sequences, we got useful results. For a finished singing performance, we would want better synchronization.

The song and video brief

Flow Music supports song creation and editing and native music-video generation. We used it to create Let the Morning In, an indie-pop track with a female lead vocal and these original opening lyrics:

I open the window, let the morning in
Shake off the shadows, feel the light again
Step through the doorway, leave the night behind
I open the window, let the morning in

The initial video brief requested a woman singing to camera, then opening a window. We used landscape 16:9 at 720p, supplied a character reference and disabled text overlays.

First attempt: the reference pose carried through

The first output showed a woman sitting with a mug, followed by a window cutaway. It did not show the requested singing performance or her opening the window.

Our reference image also depicted a seated woman with a mug, so it was poorly matched to the intended action. We corrected that input in later attempts.

Our chat confirmation became the video's displayed prompt

Flow Music's proposal described a woman singing to camera and opening a window. We approved that proposal with this chat reply:

The completed video's Prompt field then showed that same reply, word for word, instead of the detailed singing and window-opening direction:

The visible record shows our approval message being used as the displayed video prompt. That could help explain the missing performance, but it does not establish whether the model received additional instructions or what caused the failure.

Follow-up: singing appeared, but the timing needed work

We changed the reference to show the woman standing beside a window, simplified the task to singing, and repeated the complete direction in the final confirmation:

One uninterrupted close-up. She visibly sings the opening lyric directly to camera, with active lip and jaw movement. No mug, drinking or cutaways.

The results showed expressive mouth movement and a recognizable performer, but our playback review found timing mismatches with the vocals.

A longer attempt produced multiple performance shots. It still needed synchronization work for our intended use.

Saved durations include branded outros. We changed both the reference and instructions between attempts, so we cannot attribute the improvement to either change alone.

Credits: some affordable-looking requests were rejected

Completed jobs' balance changes matched their quoted costs. However, other requests were rejected even though their estimates were below the displayed balance:

In the rejected 34-second attempt, the interface displayed “Not enough credits” while the assistant immediately claimed generation had started. This occurred before the successful 30-second render in the same conversation.

Google's subscription guide says core features are available at no charge; paid plans add credits, generations and access to the latest models. The pricing page lists a Free plan, but the pages we checked do not explain the rejection message's ‘Members get one free video’ allowance. We could not determine which credit or eligibility rule caused these rejections.

Can it cover a full song?

Thirty seconds was the longest request this account accepted, not an established product-wide limit. The onboarding advertised videos up to two minutes. When we requested our complete 155.1-second song, Flow Music proposed two parts:

For this 2-minute-35-second song, the proposed workflow required two renders and stitching them together. The 1,939-credit total was quoted, not tested; we did not complete a full-song video.

Describe the video you want…

Pixo

Create AI videos with Pixo

Turn any idea into a publish-worthy video. One sentence is all it takes.

Start Creating

What we would use it for

Flow Music produced original tracks, reusable character references and downloadable videos within one application.

For synchronized singing, our examples needed more work. If that is your goal, test a short section with clear vocals and a suitable reference first. Compare the mouth movements with the actual words before committing to a longer production. If a request is rejected despite an apparently sufficient balance, clarify the credit rules before assuming a top-up will resolve it.