Sonny can now produce videos for your social posts. Ask him for a video in chat and he prepares everything: the script, the visuals, and the narration. You review it, click Generate, and a few minutes later your video is ready to post.
🎬 How it works
Ask Sonny for a video the same way you ask for a post. For example: "Make a 15-second video introducing our new collection."
Sonny creates a video project: a card right in the chat that holds the script, the attached references, and the settings (format, length, subtitles, quality). Nothing is generated yet: you stay in control.
Review the card. Read the script, adjust anything you want, or ask Sonny to change it for you.
Click Generate. It takes a few minutes, and you can keep chatting with Sonny while it runs.
Compare takes. Every generation is a take. "Generate again" adds a new take without deleting the previous ones, so you can pick your favorite from the Takes list.
🖼️ Star in your own videos
To feature real things (you, your product, your app), attach pictures to the video project as references. Click the add-reference button on the card to upload files or pick from the images you already shared with Sonny.
Sonny mentions each reference in the script (like @Image1), so the video knows exactly where each asset appears. You can attach up to 30 images, 10 video clips, and 10 audio clips.
For a video that features you, use a sharp, well-lit photo of yourself, at least 300 pixels on each side.
🎙️ Make it speak with your voice
Want the narration to sound like you? In the reference picker, click Record and speak for a few seconds (between 2 and 30). The video's narration will imitate the voice in your clip.
No recording handy? Describe the voice in the script instead (tone, pitch, accent, what makes it unique) and the video will follow your description.
💬 Subtitles
Turn subtitles on or off from the card before you generate. Three styles are available:
Clean: simple, readable white text.
Karaoke: bold words that light up as they are spoken.
Bold: heavy uppercase statements.
Subtitles are burned into the video itself, so they play everywhere, no platform settings needed. That also means they are permanent: to change or remove them, change the setting and generate a new take. If a video has no narration, it simply ships without subtitles.
⚙️ Pick your format
Aspect ratio: 9:16 for Reels, TikTok, and Stories (the default), 3:4 for the Instagram feed, 1:1 for square posts, 16:9 for YouTube.
Length: 4 to 30 seconds, or choose Automatic and the video fits itself to the script.
Quality: SD looks great on social feeds. Switch to HD when you want extra sharpness.
✨ How to get great results
💡 Attach a reference for anything real. If your product, face, or app appears in the video without a reference image, the AI will invent its own version of it.
Hook in the first 3 seconds. Viewers decide instantly whether to keep watching, so open with your strongest line or visual.
Match words to length. Narration runs at roughly 3 words per second, so a 15-second video fits about 45 spoken words. If the script is too long for the length, the video will feel rushed.
Describe what IS in frame. Write "a clean desk with a single laptop" rather than "no clutter". The video follows positive descriptions much better.
Use references for on-screen text. AI videos often misspell text they paint themselves. If exact words must appear (a screenshot, a label, a sign), attach an image that contains that text.
Iterate one change at a time. Tweak the script or a reference, generate again, and compare takes side by side.
🚀 From take to post
Happy with a take? Click Use in a post and Sonny drafts a post with the video attached, ready for your review. You can also Download the video to use anywhere else.
You don't need a connected social account to create videos, only to publish them.
Your first two videos are on us. After that, video generation uses your plan's monthly capacity. The card always shows an estimate before you generate, and a failed generation never costs you anything. See Marblism Pricing Plans & Billing Cadence for how capacity works.