This guide shows how to use Veo 3.1 when the clip needs a spoken line and a matching sound bed. You will pick text or an image, write an 8-second prompt, set the options on the current KreadoAI screen, and check whether the line, the voice, and the picture agree.
Try Veo 3.1 for FreeThe steps follow the Veo 3.1 page, including the published settings image, checked on 15 October 2026. The signed-in generator opens behind a login, so this guide uses the controls visible on that page. If your screen shows a different control, follow the control you see.
What You Can Create with Veo 3.1
Veo 3.1 text to video starts from a scene description. You name the place, the person or product, the action, and the camera. The generator builds the clip from that text.
Veo 3.1 image to video starts from a reference image. The prompt says how that person, product, or place should move, so the clip continues the look of the still.
Veo 3.1 dialogue sits in the same prompt. Name who speaks, the exact words, the tone, and when the line starts. Sound sits there too: room tone, an action sound, and whether the clip should stay free of music. Generate audio, covered in the settings step, is what lets that prompt become part of the render.
How to Generate a Veo 3.1 Video Step by Step
Choose a Generation Mode
Use text-to-video when you are inventing the scene from words. Use image-to-video when a still should stay recognizable and the prompt should describe the motion.
The published page splits those paths into Text-to-Video and Image-to-Video. Open the generator when you are ready to run a clip.
Write a Clear Video Prompt
Split the prompt into scene, action, dialogue, sound, and camera. Keep one action inside the 8 seconds, and keep one speaker on the line. The next section is a block you can copy.
Set Available Generation Options
The published settings image shows Resolution at 1080P, Video Ratio at 16:9 and 9:16, Generation Time at 8S, and Generate audio. On the image path it also shows Select image and a First & Last Frame Mode switch.
Turn Generate audio on when the prompt includes a spoken line or a specific sound. Leave the length at 8 seconds. Pick 16:9 for a wide scene and 9:16 for a vertical scene.
The words, the tone, and the timing belong in the prompt. Use First & Last Frame Mode only when you already have a start frame and an end frame. A single reference image is the right start when you have one still and the prompt carries the motion.
Generate and Preview Your Video
Generate the clip, then play the preview before you download. Compare the picture, the spoken line, and the sound with the prompt. The checklist later in this guide is the pass to use on that preview.
How to Write Dialogue for an 8-Second Veo 3.1 Video
Copy this structure. Keep each part short.
- Scene: where the person or product is.
- Action: the one thing that happens in 8 seconds.
- Dialogue: who speaks, the exact sentence, the tone, and when the line starts.
- Sound: the room or action sound, and a clear note if you want no music.
- Camera: one move that follows the action.
A sentence that needs more than a few seconds of speech gets rushed or cut. One short sentence fits the length. Two speakers in the same 8 seconds make it unclear who is talking. Music, a narrator, and a loud action at the same time compete with the line.
Example 1 is a handoff with one line.
- Scene: Daylight bike-repair shop, wooden counter.
- Action: A mechanic in a navy apron slides one matte black helmet across the counter into the customer's hands.
- Dialogue: The mechanic, calm, at two seconds, says only: "This one fits."
- Sound: Helmet on wood and quiet shop room. No music.
- Camera: Eye level, slight push-in.
Example 2 ties the line to the action.
- Scene: Quiet library desk.
- Action: A librarian places one closed hardcover into an open canvas tote, and the bag settles.
- Dialogue: The librarian, once, as the book enters the bag, says only: "Due Friday."
- Sound: Paper and a quiet room. No music.
- Camera: Slow move along the desk.
A weaker library prompt would ask two people to discuss the due date at length while music plays and traffic is loud outside. That version has no single speaker, too many words, and sounds that fight each other. The short version keeps one speaker, one sentence, and one quiet room.
The clips below are KreadoAI renders of those two shots, made to show the 8-second action. They are sample renders of the prompt. Run the same prompt on the Veo 3.1 page and judge speech, mouth timing, and sound on that preview. Picture checks below come from frames at about 1, 4, and 7 seconds. The spoken line was not confirmed by listening.
The helmet render shows a mechanic in a navy apron setting a matte black helmet on a wooden counter. By the end, a second pair of hands reaches in to take it.
The library render shows hands bringing a book to a canvas tote on a library table. The tote already holds other books before that book arrives. That extra stack is the kind of drift the checklist below is meant to catch.
Text-to-Video vs. Image-to-Video in Veo 3.1
| Task | Use this |
|---|---|
| You are inventing the scene from scratch | Text-to-video |
| You need the clip to keep the look of a reference still | Image-to-video |
| A specific person should say a specific line | Name the speaker, the sentence, and the timing in the prompt |
| You need a specific room or action sound | Describe that sound, turn Generate audio on, and check the preview |
Mode, prompt, 8 seconds, ratio, and preview stay in place. The input is what changes: words alone, or a still plus words for the motion.
How to Check and Improve Your Results
Play the preview and answer these in order.
- Did the right person say the full sentence?
- Do the voice, the mouth, and the action land in roughly the same moment?
- Does the room or action sound match the picture, and is there a sound you did not ask for?
- Do the product, the person, and the place stay stable from the first moment to the last?
- If one item failed, change that item on the next run. Change the sentence when the line is wrong. Change the sound line when the audio is wrong. Change the image when the subject drifts.
Generative video varies from run to run. Editing the prompt can steer the next render, and it can still miss the same spot.
Veo 3.1 Troubleshooting
Why is my Veo 3.1 dialogue incomplete?
The sentence is longer than 8 seconds can hold, or the prompt never names who speaks. Cut to one short sentence, and name the speaker and the moment the line starts.
Why doesn't the generated audio match the scene?
The prompt may ask for music, a voice, and a loud action together, or Generate audio may be off. Keep one sound bed plus the line, and confirm Generate audio is on before the next render.
Should I use text-to-video or image-to-video?
Use text when the scene starts from a description. Use an image when the subject already has a look you need to keep.
When should I try generating the video again?
Render again when one check fails and you have changed that part of the prompt or the input image. When the preview already matches the line, the sound, and the action, download it.
Which guide should you read next?
When the clip does not need a planned spoken line, and you want the basic text and image flow, Learn the basic Veo 3 workflow.
