Release Notes - August 2026

Julia Enthoven -

August 2026 was a big month for making Kapwing AI more capable, more controllable, and better suited to complex creative work. We launched autonomous mode for Kai, added a clearer way to reference specific media in prompts, introduced richer subtitle styles and animations, and shipped improvements across AI creation, playback, and exports. Here's more info about the updates and our motivations.

New Subtitle Styles & Animations

This month, we updated our subtitle presets with 12 new styles like Editorial, Scribble and Zing. Each preset gives you a polished starting point while keeping the subtitle text editable in the Studio.

Also, subtitle styles now change the character length per subtitle depending on the aspect ratio of the video project; vertical videos have shorter, bolder subtitles while landscape videos have longer sentences on screen.

We also added four new animations for subtitle layers within Kapwing:

  • Growing: The captions grow over the duration of a single line. Use the slider to change the size of the ending caption, making it expand faster and more or less dramatically.
  • Vibrate: Adds a subtle shake effect to the text layers.
  • Typing: A letter-by-letter typewriter style reveal for your captions.
  • Word Fly: Letters fly in and out one word at a time

After you auto-generate subtitles, click on the subtitle group and find "Growing" and "Vibrate" under the Line Animations and "Typing" under the Word Animations.

Text Layer Improvments: Bullets, Fonts, and Colors

We added support for numered and unordered lists for text layers. Highlight the text that you want to style and click the list or bullets toggles next to the font styles.

July's text refactor makes it easier to have different typography and font styles within a text layer. We plan to bring this to subtitles also soon.

Autonomous Mode in Kai

The agent became smarter, able to build on it's past steps to make a higher quality generation. One element of this change is that we replaced the "Planning" stage with a task checklist. Users see progressive updates on the video canvas as the agent makes progress and can interrupt it if they change their mind about it, recouping any unused credits.

Creators still have visibility and control over the credits expenditure as they approve the estimated credits before proceding.

The new autonomous mode means that Kai is able to handle multi-step image-to-video requests more consistently. In August, these changes to Kai led the "thumbs down" rate for Kai generations to plummet, the rate of downloading Kai generations increase, and our overall weekly-active Kai users go up. Let us know what you think about Kai.

Together, these updates make Kai more useful for building the first version of a complete project, not just generating an isolated asset.

Reference Assets in Kai

When writing a prompt, users can use the # button at the bottom of the prompt box to refer to specific filenames.

Want to define which assets are BRoll vs ARoll? Specify the ordering in your montage? Make detailed edits to one asset? Mention the filename in your prompt for better results.

New Generative Models: Seedance 2.5, Minimax H3, and Grok

Summer 2026 has been a big season for new generative AI models coming to the scene. This month, we launched new fronteir models within Kai:

  • Seedance 2.5: The most realistic AI video generation model yet. This model by Bytedance can make videos up to 30 seconds long and accepts reference images.
  • Minimax H3: The world's premiere open-weights model, Hailuo 3 by Minimax became Kai's default video generation model because the cost vs quality tradeoff seems like the best that we've seen.

We also aded the model selector to the Kai input box on the Workspace homepage, making it more visibile to all Kai users.

Recreate Individual Assets

Now, use the "Recreate with Kai" button in the right-click menu to see the prompt used to generate an asset, tweak, and regenerate. In the Studio, re-create will open the AI tab with the original prompt and reference assets.

This can also be helpful for users who are learning to write great prompts for generative media.

Support for Video and Audio References

Now, creators can reference video, audio, or image assets when they generate video, for models that support multimedia references like Seedance 2.5 and Minimax H3. This unlocks use cases like:

  • Inpainting
  • Generate with a reference style
  • Merge assets together with generative transitions
  • Making images talk

...and more. Try it out by using the "Attach Media" button to reference an asset between 4-30 seconds like and mentioning it in the Kai prompt.

More Self-Aware Skills

We empowered Kai with knowlege about it's own capabilties. Now, it is less likely to overpromise and is more straightforward with users about what it can and cannot do. If a user prompts for something that Kai knows is a limitaiton, Kai will flag that limitation during execution or pause execution for major blockers.

Faceswaps of real people are not allowed in Kai, which it highlights to the user

Kai will also validate the output after generating, self-correcting errors that it made during execution. As a result, it is less likely to report that an edit succeeded when it did not. It also is now less likely to promise users refunds or manage subscription requests (which should be routed instead through Contact Us).

Monetization Changes

This month, we made several changes to improve the Kapwing experience for free users. Our goal is to help users experience delight with Kai and retain customers for the long-term by helping them test out the product and make a more informed purchase decision. To this end, we are giving more to Kapwing free users:

  • Kapwing free users can now export videos up to 4 minutes long. Previously, the duration limit for free exports was 1 minute.
  • Kapwing users on the free plan now have 50 AI Credits to test out the product before they must upgarde.

We made corresponding changes to the pricing page as a result.

Other Changes and Bug Fixes

  • Deprecated WAN due to lack of support from our vendor
  • Increased the timeout for Upscale Video and Denoise, reducing faliures on this long-running task
  • Kai responds to requests for text, like to get the transcript or text for a video or to generate a video script
  • Fixed bugs with autonomous agent processing pasted links
  • Boost max tokens per minute to address rate limiting concerns from vendors
  • Blocked users from abusing free signups to gain more AI credits
  • Fixed an issue where users were charged credits for unfinished generations

Thanks for using Kapwing! This summer, we have revitalized Kapwing's X, Facebook, and Instagram page, so connect with us there with your questions, ideas, and feature requests.