In the process of creating these clips I stumbled into a bunch of bugs and setup needed to make it work reliably. The list of features I wanted for better control and feedback kept growing and I needed to sign up for a bunch of services to get music, voiceovers and other media into my project. I couldn't find an existing service that had everything I wanted. Seeing these hurdles and that the technology was only easily accessible to people already confident with coding agents, I started working on Klistra, and a couple of months later I'm turning it into a service so you all can use it too.
The goal with Klistra was to include everything I missed in the manual workflow into an easy to use service with a nice UI, so people could focus on what they wanted the video to look like, not setting up the framework behind it. Klistra includes precise timeline control, media generation, built in support to make animations match the rhythm of the background music, word extraction from voiceovers or songs, a dedicated iteration / feedback tool, web scraper and much more.
Something I knew I needed to get right was making it easy to iterate on small features of the video with precise control of what was being built. I designed Klistra so you first create a specification where you define scenes in detail, describe what you want shown and when, reference media or timestamps in-line, freely drag the timeline around and preview the result before letting a builder agent create the video. If you want to change something, the specification changes and the builder agent gets a diff of what changed. This way there is always a written specification that matches the render / build. There is also a planner agent built into Klistra that helps create and edit the specification for you (which is what I use the most), while letting you turn to manual fine-tuning when needed.
Don't think of this as a replacement for diffusion models, if you want photorealistic images or videos they clearly do a better job. Klistra really shines when you need motion design and precise control, exact text with your fonts, media such as logos or voiceovers, animated UI that matches your apps and so on. Klistra does have built in integrations for generating media from diffusion models, and combining the two technologies usually gets the best results.
Everything that costs money to run on Klistra, from machines to AI tokens, is paid for with credits included in a subscription or bought via one time packs (Pro starting at $25/mo and packs at $12). Being a solo-dev without funding, the budget wouldn't allow for a free tier with enough monthly credits for a real project. I instead decided to give the first 200 of you who sign up 350 free credits, enough for a simple video (the advanced tutorial used around 250 credits).
I created tutorials covering the basics. Mostly relying on the planner: https://klistra.ai/docs/basic and for advanced features and manual control: https://klistra.ai/docs/advanced
You can see demo videos on the home page https://klistra.ai/ and on the tutorial pages. I would love to see what you make!
Some additional notes:
- After feedback from friends there is also a Claude Code integration in paid subscriptions, where your agent can act as a planner, convenient if you already have a project built with Claude that has all the context and you need an explainer, ad or social post. It also reduces credit spending if you're a heavy planner user.
- Klistra can create images too.
Happy to hear what you think!