Flow AI features, one by one
Seven capabilities, what each is actually for, and when to reach for it. Ingredients to Video is the one that matters most, and the one most often confused with Scenebuilder.
Flowain · Last reviewed 2026-09-03
The feature set in one table
What Flow provides, and which layer of the stack provides it.
Every feature, and when to use it
Ordered by how much difference each one makes to the work.
Ingredients to Video
The real answer to character consistency, and the feature most people miss.
Save a character, an object, or an avatar of yourself, then reference it in any prompt by typing @ and its name. Google describes this as the most reliable way to use the same character and key objects from one shot to the next, and that matches how it behaves in practice.
The alternative most people try first is describing the same person again in a new prompt. That produces a different person every time, because a fresh generation has no memory of the last one. Ingredients is the difference between a set of clips and a scene with one character in it.
When to use it. Any time a person, product, or object needs to appear in more than one shot. Supported on Veo 3.1 Lite and Fast at 8 seconds, and on Gemini Omni Flash at 4 to 10 seconds.
Frames to Video
Supply the endpoints; Flow generates the motion between them.
You can give Flow a first frame, or a first and a last frame. With both fixed, the model is only solving for the movement connecting them rather than inventing a composition, which means fewer unknowns and fewer retries.
Since Flow now generates images itself on Nano Banana, you can make those frames in the same tool and use them immediately, without exporting to another app and back.
When to use it. Whenever you already know the composition, or when three text attempts have failed to land it. Switching mode beats rewording a fourth time.
Native audio
Veo 3.1 generates sound in the same pass as the picture.
Ambient sound, effects, and dialogue are produced alongside the video rather than added afterwards. Because they come from the same generation, they are timed to the action, which is the genuinely difficult part of scoring generated footage by hand.
Dialogue works by putting the spoken line in quotation marks inside the prompt. Short lines land better than long ones, since much past a dozen words gets rushed to fit the clip.
When to use it. It is part of Veo 3.1 generation, so it applies to every Veo clip you make.
Scenebuilder
Where a pile of clips becomes a sequence.
Scenebuilder arranges your finished clips in order, lets you reorder them, trims the head and tail of each with handles, previews the whole sequence, and downloads the result. It is an assembly tool.
It is worth being precise about this, because Scenebuilder is widely described as the feature that keeps characters consistent. It is not. Consistency comes from ingredients; Scenebuilder is what you use once the shots already exist.
When to use it. At the end, once you have clips worth keeping. For multi track editing, grading, or titles you still need a real editor.
Extend
Append more footage to an existing Veo clip.
Extend continues a clip you already generated, describing how the action should carry on rather than starting over. It is the right tool when one shot simply needs to run longer.
The constraint catches people out: Extend works only on Veo 3.1 Lite, and only at 8 seconds. It is not available on Fast or Quality. On Gemini Omni Flash it is listed as coming soon.
When to use it. When a single shot needs more length. If you need a different shot instead, generate a new one and carry the character with ingredients.
Video to video editing
Change footage you already have, with a text prompt.
Upload a clip of up to 60 seconds and one gigabyte, select a 10 second segment, and describe the change in words, such as shifting the lighting to a cinematic sunset. You can refine the result across up to three conversational turns.
This runs on Gemini Omni Flash only. No Veo model does video-to-video editing, which is a common source of confusion when people cannot find the option.
When to use it. For adjusting existing footage rather than generating new footage, including video you shot yourself.
Image generation and Custom Tools
Flow absorbed Whisk and ImageFX in February 2026.
Image generation and editing now happen inside Flow on the Nano Banana models: Nano Banana 2 Lite free, Nano Banana 2 on standard tiers, Nano Banana Pro on Ultra. A still you generate can be used immediately as a frame or an ingredient.
Custom Tools let you build your own workflows, with creation available from Plus upward. Google ships prebuilt ones including Storyboard Studio, Type Overlays, Video Resizer, Image Editor, and Character X-ray. The Agent, powered by Gemini, plans multi step edits and applies changes to uploaded video on instruction.
When to use it. Whenever you would otherwise leave Flow to make or fix a still image before animating it.
What Flow does not do
The gaps are as worth knowing as the features, especially before you subscribe.
Assembly only, no real timeline
Scenebuilder orders and trims clips, but there are no tracks, keyframes, or grading. Anything beyond a straight cut goes to a real editor.
Extend is Lite-only
Veo 3.1 gives 4, 6, or 8 seconds; Omni Flash up to 10. Extend only works on Veo 3.1 Lite at 8 seconds, not on Fast or Quality, which surprises most people.
Unreliable text in frame
Signage, labels, and captions still come out wrong often enough that you should plan to add them afterwards.
Not deterministic
The same prompt twice gives you two different clips. Scenebuilder helps continuity within a project but does not give you reproducibility.
Hands and crowds
Complex hand movement and background faces remain the most common visible failures. Framing around them is more reliable than regenerating.
Free, but metered
50 free credits a day is two Fast generations. Every attempt bills, including discarded ones, so heavy iteration pushes you onto a paid plan faster than people expect.
Pricing and credit detail is covered on what is Flow AI, and the workflow these features sit inside is on the how to use Flow AI page. To see these features written into prompts you can copy, with the mode each one needs, read the Flow AI shot prompt examples. For how the same feature set fits a design or previsualization pipeline, see Flow AI for designers.
Questions about Flow AI features
What people ask when comparing Flow against other video tools.
Which Flow feature matters most?+
Ingredients to Video. Keeping one character consistent across separate shots is the hard problem in AI video, and saved ingredients referenced with @name are Google's answer to it. Scenebuilder gets more attention but only assembles clips you have already made, so test ingredients first when comparing Flow against alternatives.
Does Flow have a timeline or editing tools?+
Partly. Scenebuilder arranges clips in order, lets you reorder them, trims each with handles, previews the sequence, and exports it. Gemini Omni Flash can also edit uploaded footage from a text prompt. What is missing is multi track editing, keyframing, and colour grading, so Flow handles assembly and your editor handles finishing.
What resolution does Flow output?+
Veo 3.1 generates 4, 6, and 8 second clips; Gemini Omni Flash reaches 10 seconds at 360p or 720p. Upscaling is a separate step: 1080p is free on Plus, Pro, and Ultra and unavailable on the free tier, while 4K costs 50 credits and is Ultra only. Leave it until a sequence is locked.
What is Flow Agent?+
A Gemini powered assistant inside Flow that plans multi step edits and keeps memory of what is in your project across a session. It rolled out globally to all Flow users after Google I/O 2026. It is an assistant layer rather than a generation feature, so it changes how you work rather than what the model can produce.
Can Flow edit video I already have?+
Yes, but only on Gemini Omni Flash. Upload a clip up to 60 seconds and 1GB, select a 10 second segment, describe the change in text, and refine across up to three turns. No Veo model supports video-to-video editing, which is why the option disappears when you have a Veo model selected.
Flowain
Writer and editor, FlowAIFX
Flowain writes and maintains FlowAIFX
Every page is checked against Google's published specifications for Flow and Veo 3.1 on the date shown, and corrected when Google changes them. Where a limit is undocumented or a credit cost is easy to misjudge, it is written down rather than left out.
Last reviewed 2026-09-03 · Report a correction
Test Scenebuilder first
Generate one shot, extend it once, and look at whether the character and the light survive the join. That single test tells you whether Flow is worth the subscription for the work you do.