Video, speech, music and finishing

Prepare references, then review the delivered picture and sound.

Updated 2026-10-22

Wavemaker makes video, speech, music and sound effects in a conversation. You describe the finished piece, the agent plans it, makes the shots and sound, cuts them together and delivers the file. This guide covers the whole path, from choosing a format to reviewing what was delivered.

These are the stages the progress line shows under your message for a video. An audio-only piece shows Brief and plan, Casting, Voice and mix and Edit and finish.

Make your first video

1. Open New chat and choose a starter, such as Create a UGC ad or Make a product launch film, or write your own brief.
2. Open the format chip beside the brief (it reads 16:9 · any length until you change it) and choose the Frame, Length and Resolution.
3. Describe the piece: the subject, audience, where it will run, the exact spoken words and the ending.
4. Select + › Upload files for product photos, footage, a logo or a song, or Add from Library to reuse something you already have.
5. To approve the work in stages, ask for a script or shot list first and end your brief with "Wait for my go-ahead". Leave Step by step off: the web app can't approve its steps yet. See Plan in stages.
6. Open the quality menu beside Send (it shows Balanced by default). Under Planning budget, choose Check what this message costs to see the expected cost before you send. Checking is free.
7. Press Send, then follow the progress line. Review the draft cut, ask for changes, then ask for the final.

Make a 20-second vertical UGC ad for our cold brew. A creator talks to camera about why she drinks it before work, with b-roll of the can and captions. End on the line "Cold brew, ready when you are."

Check that it worked

• The conversation's Saved plan view shows each request as Delivered or Partly delivered once files were delivered. A finished file that was never delivered does not count; see When work counts as delivered.
• Under the delivered result, check chips show what passed, for example the duration, the frame size, loudness in LUFS, Spoken words match the script or Text inside safe margins. A failed check never shows as a chip. When the spoken words differ from the script, Spoken words differ from the script appears under Needs a look.
• A note reading Not checked by ear. or Spoken words were not checked against the script. means nothing listened to that part. Listen yourself before you publish it.

Format: type, frame, length and resolution

The format chip records what the message asks for without changing your words. A type becomes a video the reply is asked to deliver at that frame, length and size. A frame or resolution without a type is passed to the agent as a note.

Every type also offers Any length. Choosing the selected type again clears it.

Resolution is Auto, 720p, 1080p or 4K. The resolution names the short side of the frame, so 1080p vertical is 1080 × 1920. A frame wider than 16:9 keeps the 16:9 width: 1080p cinema is 1920 × 804.

How shots are made

By default the agent makes and reviews still frames first, then animates them. Stills let it hold faces, products, clothing and framing steady before any video is spent. To animate a still you already like, open its menu and choose Animate this still. That adds the still to your brief as a target; nothing is made until you press Send.

The default video model, H3, can start a shot from:

• a text description only;
• a first frame (a reviewed still);
• a first frame and a last frame, which fixes where the shot starts and ends;
• reference images of the people or objects in it.

For on-camera speech, the agent by default records the line first and uses it to guide the shot, one speaker per shot, so the lips are made in sync. A longer piece is cut from many short takes, so a 30-second film can need many shots.

Video models

Unless you choose a video model, video starts on H3 · self-hosted, Wavemaker's own video model, at every quality level. Partner video models run when you choose one or name one in your brief. If you or your workspace saved different default models, those are the starting choice instead.

To choose, open the quality menu, select Show model choices, and set the Video row. The list shows which models you can use now; a model that cannot run for you is marked Unavailable. Partner models in the list include Gemini Omni 1.1 Flash, Kling 3.0, Runway Gen-4.5, Luma Ray 3.2, Seedance 2, Grok Imagine Video, MiniMax H3 (partner API) and Vidu Q3. LTX 2.5 · self-hosted is another self-hosted option.

You can also pick a saved preset from the same menu:

Choosing a preset also sets its quality level: Balanced for Standard, Draft for Draft and High quality for the two partner presets, which are labeled Partner video · higher spend. A model you choose applies to your next message. When you choose one, work that would use a different model for that kind is refused rather than swapped. When you leave video unchosen and an H3 shot is refused or keeps failing, the agent can move that shot to a partner model, and it tells you which shots moved and why.

If you saved your own key for a provider in Connections, that provider's models run on your key and your provider bills you. See Credits.

Speech, music and sound effects

The Voice row includes ElevenLabs (the default), OpenAI speech and MiniMax speech. The Music & sound row includes ElevenLabs music & sound and ACE-Step music · self-hosted; leave it on Auto and the agent chooses.

• Speech: supply the exact words, and spell names and unusual words the way they should sound. To design, clone or reuse a voice, see Create speech and reuse a voice.
• Music: describe the mood, tempo and where it should build or stop, or upload your own track. A music video can run the length of your song.
• Sound effects: name the sounds and where they fall, for example "a can opening at 0:02".

The agent can measure the mix, the loudness and the spoken words against your script before it delivers.

Draft cut and final render

A draft cut is a quick, cheaper render of the same edit at about a 540-pixel short side, with the same timeline, audio and graphics. Use it to check timing and order.

• Each render is its own charge. A draft is never turned into the final; the final is rendered again at full size.
• A draft can never be delivered. Only a final can.
• The agent does not have to draft first. To check a draft cut before the final render, ask for a draft in your brief and end with "Wait for my go-ahead".

The Draft quality level is a different setting: it asks for fewer takes and lighter checks across the whole message.

Captions and on-screen text

Say which words need captions and where the video will run. Burned-in captions sit in a stable band near the bottom of the frame, kept apart from titles and the call to action. On 9:16 video, captions sit above the bottom 20% and below the top 14% of the frame by default, clear of app controls; with the TikTok safe zone they sit above the bottom 26%.

You can also ask for caption files to go with the delivered video, as SRT or WebVTT. They are made from a transcript of the finished audio: at most two lines of 42 characters per caption, each on screen for 1 to 6 seconds. The transcript is charged for the length of the audio.

Edit the timeline yourself

Each conversation keeps its cut as a timeline, and the next render uses what is saved there.

1. Open ••• › Edit timeline in the conversation.
2. Drag the edge of a clip to trim it, or drag the clip to move it. Use Video track and Audio track to add tracks.
3. Choose Save timeline. Each save becomes a numbered revision.

With a clip selected, [ and ] trim it (Shift for 1 second, Alt to extend), Alt with the arrow keys moves it, and Delete removes it. ⌘/Ctrl Z undoes and ⌘/Ctrl S saves.

Saving does not render anything. Ask the agent for a new draft or final when you are ready. If someone else saved while you were editing, you see The timeline changed and choose Keep my edit or Use the saved version; nothing is overwritten without asking.

When work counts as delivered

A file is delivered only when the agent delivers it as the finished video, audio, image or set for your request. A finished take, a draft or a file in Outputs is not a delivery.

When a delivery names a destination (say where the video will run, and the agent names it), the delivered file is checked against that destination's specification. A file that misses it is refused, and the agent must fix it before it counts as delivered. Every video destination checks the aspect ratio and minimum size, the duration, H.264 video at a constant frame rate with square pixels, AAC audio, and any file size cap. Pinterest also accepts H.265.

Audio ads can go to Spotify (MP3, WAV or OGG at 44.1 kHz, up to 30 s) or general streaming audio (up to 60 s). Loudness is measured and reported against each destination's reference, but it never blocks a delivery. Without a destination, nothing beyond the file being yours and a final render is checked.

In the Saved plan view, each request shows one of these states:

You can also deliver a take yourself: in Saved plan, choose Choose this take, then Deliver as the video (or audio, image or display set). Keep as a project asset (not a delivery) saves it without delivering it.

Add your brand watermark

A watermark always makes a new copy of the video with your logo burned in. The original stays clean.

1. Open Settings › Brand and watermark and add your logo: PNG, WebP or JPEG up to 25 MB, or a plain SVG up to 2 MB. Animated images are not accepted. You can save a separate logo for dark footage.
2. Set the placement: Position (Automatic is bottom right, or top left on vertical video), Keep clear of, size as a percentage of the short edge (3–30%, default 10%), When it shows, Legibility, and opacity (default 0.6) and fade (default 0.4 s). Choose Save watermark settings.
3. To watermark one video, open its menu and choose Add watermark, check the preview, and choose Add watermark. The request is added to your brief; press Send to make the copy.

Turn on Watermark new finished videos automatically and each message you send in the web app from then on asks the agent to deliver every finished video twice: clean, and with your logo. Work started from the API, an MCP client or the CLI doesn't get this request. The watermarked copy is a new render and is charged like one. If you saved a separate logo for dark footage, the agent picks the light or dark version to suit the footage.

The free-plan badge

Until the workspace has paid for credits or a plan, every video Wavemaker generates for it is saved with a small badge burned in: the Wavemaker icon and "created with wavemaker.io".

• It sits in the bottom-right corner of landscape and square video. On vertical and 4:5 portrait video it sits at the top left, 14% down from the top edge, below the area apps use for their top bar.
• It is 85% opaque and 7.5% of the frame's short edge tall.
• A cut made from badged clips shows the badge too. The agent delivers it as usual and does not crop, cover or remove it.
• No clean copy is kept. Videos made after the workspace's first payment have no badge, but paying does not remove the badge from earlier videos.
• Stills are never badged, and neither is video made on your own provider key.
• If the badge cannot be applied, that video fails instead of being saved without it.

See Plans to upgrade.

Review picture and sound

Play the whole delivered file with sound, at full size, before you download or publish it. Check:

• every spoken word, name and pronunciation against your script;
• lip sync on speaking shots;
• faces, hands, products and spelling in key frames;
• timing, transitions, the ending and the call to action;
• music and voice levels, silences and clipping.

To change one part, name the shot or time range and say what to keep: "Keep shots 1 to 3. Make the last shot hold for two more seconds." Use Compare with version N and See the other takes in a result's menu to compare before you decide.

If a shot or render fails

The progress line lists each shot as Waiting, In progress or Didn't finish, with its takes. When a shot fails, you see Shot N didn't render and three choices: Try shot N again, Use another video model or Skip it and finish. Shots that finished are kept. Each choice sends a new message, which is new work and is charged. Use another video model lists the video models you can use now and also sets your Video row to the one you pick, so later messages use it too until you change it.

Each call reserves credits before it runs, and credits it does not use are released. If you lose your connection, reopen the conversation instead of sending the brief again; a repeated brief is new work and is charged again. See Recover interrupted work.

What affects the cost

• Length and number of shots. Longer pieces need more shots and more takes.
• Quality level. High quality asks for more takes and closer review; Draft asks for fewer.
• Video model. Partner presets are labeled higher spend; H3 is self-hosted.
• Renders. Every draft cut, final and watermarked copy is its own render.
• Extras. Voices, music, sound effects and caption transcripts are each charged.

Use Check what this message costs before sending, and set a Total planning limit (credits) under Planning budget to cap what one message may spend. See Credits.

Related guidance

• Understand credits, reservations and charges
• Download results and recover an unavailable file
• Create and edit an image
• Recover interrupted or uncertain work
• Choose, compare and revise a result
• Create speech and reuse a voice