Can Claude make video ads? Script yes, video via tools

Partly. Claude can write a video ad's concept, script, and shot list, but it outputs text, so it can't render the video itself. With video, voice, music, and editing tools connected, Claude can call them to make the clips, the voiceover, and the music, and assemble them into one ad file.

Claude's side comes from Anthropic's models overview and custom connectors article. The tools are Sume's hosted MCP server, per MCP tools and gates, Video generation, and Timeline 1.0. All were read on 2026-09-29. For the brief-and-approve workflow in general, see How to make a video ad with AI.

What does Claude do, and what do the tools do?

Anthropic says all current Claude models take text and image input and give text output. Claude's part is the thinking: the hook, the script, one prompt per shot, and the order of the tool calls. Sume has no official Claude connector; you add https://mcp.sume.com/mcp as a custom connector (Free Claude plans get one) and turn Write on at Sume's consent page. Sume's basics page says hosted MCP still works but is not the primary path today.

• generate_video renders each shot from a prompt, or from an approved still.
• tts_create reads the script as a voiceover.
• music_create makes a music bed. Music 1.0 is retiring gradually; its requests now resolve through the Music Router.
• timeline_create joins the clips over the voiceover, with the music as an optional soundtrack, and returns one MP4.

How are the pieces put together?

Timeline 1.0 takes one audio spine plus up to 200 ordered video slots. The voiceover is the spine and sets the ad's length; each clip gets a start time and a duration, and later slots can take a fade or wipe transition. Every URL must already be a media.sume.com file in your workspace; completed Sume jobs return their outputs as media.sume.com artifacts, so the earlier tool calls supply them.

One trap: in current code a Timeline render takes sound only from the spine and the soundtrack, so any sound a clip was generated with is dropped. Put everything the viewer should hear in the voiceover or the music.

Can Claude make a 4:5 or 9:16 ad?

9:16 and 16:9 yes; 4:5 needs one more step. Each video model lists its own supported_aspect_ratios; the docs example for Seedance 2.0 lists 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. In current code, 4:5 is not a ratio any video model can advertise. For a 4:5 feed ad, have Timeline output the frame instead: output.width and output.height take even integers from 256 to 2160, so 1080×1350 works, and the default cover fit fills the frame by cropping. Timeline's default output is 1080×1920, so set the size for any other shape, 16:9 included.

Each ad platform publishes its own size and length rules; check them before you render. Meta video ad specs covers Meta's feed and Reels placements.

How much does a video ad from Claude cost?

Each piece bills your Sume wallet, not your Claude plan, plus a 5.5% agent fee by default. Routed video models are priced at the provider's list price × 1.25; How much does an AI video cost? shows the math. Ask Claude to send each paid call with dry_run=true first: it previews the cost without submitting the job.

What can't Claude do here?

A few limits to plan around:

• Use footage from your laptop. Hosted MCP cannot read local files, and Timeline takes only Sume-hosted media.
• Promise results. Nothing here predicts how an ad performs; test it like any other creative.
• Make one long shot. Each model lists its supported_durations, and seedance-2.5 and wan-3.0 accept up to 30 seconds per clip, so a longer ad is several shots joined on the Timeline.