When many people use Google Flow for the first time, they will be stuck with three questions: whether it is expensive, how to write Prompt, and whether the character will change faces. In fact, there is no need to study all functions at the beginning. Today's goal is only one: to generate and save a 6 - 8-second AI video within 20 minutes. Putting together multiple shots, fixed roles, and continuous shooting are all advanced contents in the future.
What is Flow?
Bottom line: Flow is Google's AI video creation workbench, not just a video generation button. which mainly comprises Veo (generate video)、Imagen / Nano Banana (generate reference maps and characters), Gemini Agent (chat, split mirror, write Prompt).
Entrance:
Newbies can remember 6 functions first:Text to Video(Text generation video),Agent(Chatting out of the camera),Characters(Fixed role),Frames(Control the front and rear screens),Scenebuilder(Put together multiple shots),Extend(Continued). The correct way to play Flow is not to generate the entire ad with one click, but:Short shots → splicing → Continue shooting if necessary.
Can I use it? How much is it?
Confirm three things first: complete age verification (recommended 18+), support Flow in your region, and check the points in the upper right corner first after logging in. Flow changed topoint system, stop referring to the previous "100 times a month" tutorial. Lite models are the cheapest and suitable for practice;Quality is reserved for final finalization; points will be deducted according to a few items generated at a time. Remember one saying: **Lite trial and error, no matter how expensive the final version is, the good film is saved first. ** In addition, the generated videos will carry SynthID, and some packages will also have visible watermarks, which is normal.
Opening it for the first time, just doing this
Log in to Flow with Chrome, confirm the points in the upper right corner, create a new Project, and find the left material area, middle results area, and bottom input box (Agent). Don't study advanced features such as Avatar and 4K today. There is only one goal:Make the first video first.
The easiest way: Use Agent directly
Open Agent and put Confirm before generating set to AlwaysAvoid mistakenly deducting points. Then just enter one sentence, such as: "Help me make a promotional video for a cozy coffee shop." If the Agent asks, he will answer: the atmosphere is "warm and lively", the selling points are "hand-brewed coffee, wooden bar, afternoon sunshine", and the style is "warm color, realistic movie feel, and shallow depth of field."
First modify the split lens generated by the Agent, and then start generating video. Recommended parameters:Video, Veo 3.1 Lite, 1 output, duration 6 - 8 seconds, 16:9 (horizontal screen) or 9:16 (vertical screen)。Save it immediately if you are satisfied; if you are not satisfied, only modify one variable at a time, such as light, mirror movement or action, and do not change three or four items at a time.
Chinese Prompt is completely okay
Don't rush to write English. Rather than "a woman walking in a coffee shop", it is more recommended to write it: "Mid-shot, a thirty-something-year-old coffee shop owner wearing a linen apron is rushing to make coffee in front of a wooden bar. The afternoon sun shines through the floor-to-ceiling windows. Inside the store, it has a warm cinematic color, a shallow depth of field, and a coffee machine and a soft chatting environment sound in the background." If you want to convert it into English that is more suitable for Veo, you can leave it to Gemini to rewrite it.
How to avoid changing faces?
Don't re-describe the characters in every shot. The correct way to do this is: Characters → Create a role → Upload a reference picture → Name it (such as Ava) → Write @Ava directly for subsequent Prompt. For example: "@Ava walks into the coffee shop, opens the glass door, and smiles at the camera." In this way, the character consistency in the second shot will be much higher.
How to spell multiple shots?
After generating two or three satisfactory videos, click Add to Scene, enter Scenebuilder Adjust the order, cut the head and tail, and use it when needed Extend Continue generating follow-up actions and finally download the entire video. Compared to direct growth videos,Stitching 2 to 3 short lenses usually has higher quality and saves more points.
common problems
The picture is very random, most of it is Prompt space, which can add characters, actions, scenes, lenses and light; if the points drop too quickly, change back to Veo 3.1 Lite and output 1 item. Don't use Quality to try and error; if the characters keep changing, use Characters or Ingredients, and then pass the @ Role Name If the generation fails, priority will be given to checking the region, age verification and content policy. Generally, no points will be deducted for failed generation.
First Hour Checklist
- □ Log in to Flow and confirm points
- □ New Project
- □ Open Agent and set up Confirm before generating = Always
- □ Enter an idea and answer the Agent's question
- □ Use Veo 3.1 Lite Generate 1 6 - 8-second video
- □ Save immediately if satisfied
- □ (Advanced) Create a permanent character for Characters
- □ (Advanced) Scenebuilder splicing multiple lenses
Remember three sentences:Flow is a workbench, not one click to produce blockbusters;Lite is responsible for trial and error, and good films are saved immediately; Agents do lens division and Characters lock roles, which is more important than writing complex English Prompt.