TikTok is giving advertisers expanded capacity for AI video generation, integrating the latest model from parent company ByteDance to produce better-looking, more engaging clips.
TikTok announced integration of the Dreamina Seedance 2.5 video generation model, saying it produces better quality clips designed specifically for the TikTok audience.
The headline change is length: advertisers can generate 30-second video clips, up from 15 seconds previously, enabling deeper storytelling. Advertisers can also upload up to 50 multimodal references — image, video and audio — significantly expanding the ability to guide generation. The previous limit was nine.
On quality, TikTok says the new model generates "sharper, more realistic imagery" with stronger preservation of lighting, motion and character consistency across a video, enabling a more advanced understanding of product context, proportions and scene composition so advertisers can execute more ambitious ideas "with a stronger finish."
TikTok integrated Dreamina 2.0 in April, and this update builds on that momentum.
The timing is awkward. Many social platforms, TikTok included, are dealing with significant user backlash against AI-generated video, with users clearly communicating that they do not want AI slop taking over their feeds.
Last week Snapchat announced it will no longer promote wholly AI-generated content in its Spotlight short video feed. Pinterest, Reddit and X have also initiated measures to limit the spread of AI-generated content. TikTok itself has implemented transparency and user education initiatives and gives users the option to indicate they want to see less AI-generated content in their feeds.
In that context, giving advertisers more ways to use AI could run counter to user sentiment and even create backlash for advertisers who lean on it. Clearly TikTok sees potential in the model regardless; it is rolling out to select paid advertisers in select markets.
Separate use cases. Product shots, backgrounds and variants are strong candidates for generation; filling the brand's human-facing creative with generated assets raises exposure on both platform policy and audience reaction.
Longer is not automatically better. Thirty seconds buys narrative room and adds drop-off points, so revisit the opening hook when you extend length.
Use the 50-reference ceiling deliberately. It is an opportunity to feed brand guidelines — logo, color, tone references — into the model for more consistent output.
For the platform-wide brakes on AI content, see Can the Apps Stop AI Slop?; for the specific recommendation policy, see Snapchat Stops Recommending AI Slop.
Generated ad clips extend from 15 to 30 seconds and multimodal reference uploads rise from 9 to 50, with improved lighting, motion and character consistency.
It is rolling out to select paid advertisers in select markets.
User backlash is rising — Snapchat, Pinterest, Reddit and X have all limited AI content — so leaning on generated assets for brand-facing creative can invite negative reaction.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
Meta has begun alerting advertisers that the Placements option is being removed from ad sets, along with platform exclusions. Another notch of control disappears into automation.
NBT signed an ad server deal with Mobwith to expand into display advertising. What advertisers should check when an 11M MAU reward network starts selling standard display inventory.
Naver display ads reach reporting launched August 12. Here is what it measures, why logged-in-only counting and 10,000-unit buckets matter, and which campaigns are excluded.