Platform-Specific AI Video Specs for Meta, TikTok, and YouTube Ads

Platform-specific AI video specs for Meta, TikTok, and YouTube ads belong in the creative brief. Dimensions, safe zones, timing, audio, and disclosure rules shape what an ad can say and how viewers receive it. Platform-native production starts with reusable workflows that preserve what each campaign teaches the team.

Why platform-specific specs are a production decision, not a post-production fix

When creative is rejected, the time spent making it is lost before an impression runs. Cropped hooks and hidden calls-to-action waste spend after approval, when the ad technically works but its message does not. A video repurposed without adjustment can also lose reach and pay more in CPM because the platform reads it as low-quality. Passing review is not the same as presenting well.

Those failures begin before the final file is made. A face, product, or headline stays visible within the placement’s interface only with suitable framing. The message must arrive before the viewer leaves, and its delivery depends on hook timing and audio strategy. Disclosure adds another constraint, with violations risking rejection or account strikes. Every one of those decisions depends on the intended platform and placement.

Meta video ad specs, placement by placement

Meta’s Reels and Stories placements on Instagram and Facebook use 9:16 vertical video. They accept MP4 and MOV files up to 4 GB. GIF is also accepted for Stories and Facebook Reels, but not Instagram Reels. Video encoding calls for a supported video codec, square pixels, a fixed frame rate, and progressive scan. Audio uses stereo AAC at sufficient bitrate.

The frame needs room for the interface. Text and logos should stay clear of the top and bottom portions and leave a margin on each side, where profile icons, ad copy, and other controls can overlap them. The hook belongs in the first 3 seconds, with timing also allowing content to clear the safe zone. Reels favors a short duration, but that creative target is separate from a hard upload limit.

Facebook and Instagram Feed also accept 1:1 square video at 1080×1080. Feed follows the same file type, encoding, and audio rules described for Reels and Stories. A minimum accepted resolution is not a recommended production target. An ad can pass technical review and still look poor in its placement.

Meta is a mixed sound-on and sound-off environment. Captions are necessary for the message to remain understandable without audio, and the opening hook must read visually. Meta Advantage+ Creative, built into Meta Ads Manager at no extra cost, automatically generates and enhances variations. Its production role extends beyond export and returns later in the workflow.

TikTok video ad specs, placement by placement

TikTok In-Feed Video Ads recommend 1080×1920, with 540×960 as the minimum accepted resolution. MP4 and MOV are accepted, with a maximum file size of 500 MB. Critical content should avoid roughly the top 130–240px and bottom 440–660px, where navigation, captions, and the CTA bar appear.

The hook window is the first 2 seconds. TikTok defaults to sound-on, so audio pacing and voice-over timing are part of the creative structure. The vertical shape may match Meta’s, but the safe zones, opening window, file-size limit, and audio expectation differ. A Meta-ready export is not automatically TikTok-ready.

Spark Ads use existing organic posts from a brand or a creator with permission. They match the original post’s ratio and duration, follow organic limits, and preserve its views, likes, and comments. Their disclosure route is different as well. AI-generated Spark Ads follow TikTok’s community AIGC rules rather than the ads-specific toggle used for standard In-Feed ads.

TopView Ads appear immediately when someone opens TikTok, full-screen without competing content. They use 9:16 video at 1080×1920, accept MP4 and MOV, and have a 500 MB file limit. Brand Takeover Ads also appear full-screen on app open, using video or static creative for a few seconds at 9:16 and 1080×1920. One advertiser per category per day makes that placement exclusive and expensive.

TikTok Symphony Creative Studio has native AI tools for making videos. Symphony Digital Avatars includes Voiceover and Product types for presenter videos, while Remix Videos generates TikTok-optimized ads from existing footage and assets. The studio automatically adds an AI label to content made with it. That helps with disclosure, but other production routes require different handling.

YouTube video ad specs, placement by placement

YouTube Skippable In-Stream Ads use 16:9 landscape as the standard, with 1920×1080 resolution. Demand Gen campaigns also support 9:16 vertical creative, following the retirement of Video Action campaigns in April 2026. There is no hard minimum or maximum length, though the platform says shorter videos are best. Google Ads Help gives a file-size limit of 256 GB. The early skip button makes the opening seconds the real constraint.

Non-Skippable In-Stream Ads use 16:9 at 1920×1080. Viewers cannot skip, so the message and CTA must fit the allotted duration. Bumper Ads impose a different kind of discipline. Their maximum is 6 seconds, a hard cap rather than a creative recommendation, and they are typically used for reach and frequency with longer formats.

YouTube Shorts Ads use 9:16, with 1080×1920 recommended. A duration under a minute is recommended. Landscape in-stream creative and vertical Shorts therefore need different compositions. The same distinction applies to disclosure placement, since a label must remain readable in the format viewers actually see.

AI content disclosure rules across Meta, TikTok, and YouTube

Each platform handles disclosure in its own way, and political advertising has stricter rules than ordinary commercial creative. That distinction needs to be settled before generating a presenter or synthetic scene. Otherwise, a finished video may need more than a new label.

Meta requires disclosure for political and social issue ads showing realistic people saying or doing things they did not, realistic-looking people who do not exist, or events that did not happen. Undisclosed ads are rejected. Meta’s own generative AI tools cannot be used for political ads. For commercial ads, Meta automatically labels significant edits and AI-generated photorealistic humans created with its tools.

Since June 1, 2026, Meta also detects and labels ads created or edited with third-party AI tools through industry-standard signals. The label appears under “About this ad” in the three-dot menu without advertiser action. Presenter framing still needs care. An AI brand spokesperson is permissible, but presenting that spokesperson as a real customer testimonial crosses into FTC fake-review territory.

TikTok requires the “This ad contains AI-generated content” toggle for non-Spark ads with completely AI-generated or significantly AI-modified images, video, or audio. The toggle locks after submission and resets when a campaign is duplicated, so every copy needs checking. Spark Ads instead use the organic AI-generated label, a sticker, or the caption. TikTok has also read C2PA Content Credentials to auto-label uploads from some platforms since May 2024. Political advertising is banned entirely.

Beauty or lighting filters, AI-assisted scripts and thumbnails, upscaling, and cloning one’s own voice do not require disclosure. Labels appear in the player for photorealistic content and in the expanded description otherwise. YouTube can apply them through its own tools, C2PA metadata, or detection.

Google Ads election advertisers must select “Altered or synthetic content” in campaign settings. Google generates disclosure for some formats, while others require visible wording such as “This video content was synthetically generated”. In July 2026, Google added an optional AI label setting for image and video creatives across all ads, noting disclosure requirements for certain ads in the EU, India, and New York. The cross-platform production lesson is to keep visible text such as “Created with AI” throughout the video while completing each platform’s required disclosure steps.

Why building to platform specs from the start changes how AI production pipelines must be structured

A single master cut makes the wrong decisions too early. Its framing, audio rhythm, and opening sequence are already fixed before the placement gets a say. Turning that cut into TikTok, Meta, and YouTube versions takes structural re-cuts as well as resized files.

Platform-native production reverses that order. Shots are generated to fit the frame and its safe zones, with room for disclosure text during the video body. TikTok In-Feed and YouTube Shorts may have the same resolution even when their shots are composed differently. Neither needs to be a crop of the other.

At volume, those choices also shape model routing. Text-to-image models handle concepts and storyboards, models strong in dialogue and character consistency handle talking-head UGC, and models strong in cinematic motion handle establishing and hero shots. The brief stays the same as the prompts change. Action and continuity may need emphasis for one model, while lighting and composition guide another.

That split avoids sending every task through one premium model. It also creates a need for a unified prompt library with model-specific annotations. Voyager describes itself as an open harness for creative work, tuned so that Claude, ChatGPT, Kimi, DeepSeek, and open models can work more effectively with creative tools. Its pre-built and custom skills fit this integration point, where a shared brief must meet different model strengths without fragmenting the production process.

How AI skills — pre-built and custom — handle platform-specific production at scale

Advantage+ Creative can animate static product shots into looping video, swap backgrounds, and generate music matched to a video’s tone. As of 2026, it also supports AI dubbing for localization. These tools work within Meta’s placement requirements, so there is less work to do by hand when making variations.

Symphony offers a similar way to get started on TikTok. Its avatars can make talking-head UGC-style content, while Remix recuts existing material into TikTok-native formats and the suite automatically adds disclosure labels to content made with it. That native fit has real value. The limitation is cross-platform use, because creative tuned to one platform’s auction and aesthetic can read oddly elsewhere.

Product-URL tools pull images, copy, and pricing into UGC-style video ads, making it easier to produce many e-commerce variants without a shoot. Performance prediction engines use data from each platform to predict which variations may work best before money is spent. These forecasts inform creative judgment during selection.

Custom skills make the process repeatable across campaigns. Prompt templates can embed safe zones and hook timing, while production logging records the model, cost, time, and prompt version for each generation. Keeping scenes within a project gives those records a shared context. Voyager’s combination of pre-built and custom skills provides a starting structure that teams can extend with prompt libraries and routing rules. The durable asset is the team’s own production logic.

How adaptive workflow memory turns per-platform production knowledge into a compounding team asset

Past production work can guide what goes into the next brief. A remembered TikTok hook can preserve the timing, audio framing, and safe-zone choices that worked without forcing the team to derive them again. Meta Reels needs its own record. Keeping those distinctions prevents one platform’s successful treatment from becoming an unquestioned default for every placement.

The same principle applies to model selection. Cost, time, and output-quality records across campaigns give a team a routing policy grounded in its own briefs and brand voice. Modular production makes that knowledge usable. Hooks can differ by platform, caption emphasis can change with audience familiarity, and intros and outros can reflect context.

Voyager’s persistent memory is designed to adapt to individual workflows so that repeated creative work becomes faster, cheaper, and higher-quality over time. The value is retaining which production choices worked for a particular audience and placement.

That changes where an editor spends attention. Automation handles repetitive work such as rough cuts, silence removal, auto-captioning, safe-zone reframing, and multi-format export. The editor’s higher-value task is deciding which of thirty generated variants to scale, reading performance signals, and updating the prompts and routing rules. Assembly becomes less of the job, leaving more room for judgment.