I've been working on LTX 2.3 for the past few weeks, focusing on AI video generation workflows.
At first, I assumed users mainly cared about generation quality.
I was wrong.
After talking with creators and testing different workflows, I found that most people were asking about three things:
Faster iteration
More predictable outputs
Simpler workflows
Instead of spending another week adding new visual effects, I decided to focus on improving the creation process itself.
What I Changed This Week
Simplified the prompt workflow
Improved generation consistency
Reduced the number of steps needed to create a video
Added more example prompts for testing
What I Learned
One interesting lesson was that users rarely judge an AI video tool by a single output.
They judge it by how quickly they can get a usable result.
A creator might generate 10–20 versions before finding the one they like. If the workflow feels complicated, they leave.
That changed how I think about product development.
Current Workflow
My current workflow with LTX 2.3 looks like this:
Write a simple scene description
Generate a first draft
Adjust motion and camera instructions
Export multiple variations
Select the strongest result
This process has been much more reliable than trying to create the perfect prompt on the first attempt.
What Surprised Me
The biggest surprise was that creators preferred consistency over complexity.
Many users told me they would rather have predictable results than dozens of advanced settings.
That feedback is influencing every feature decision going forward.
Next Week
For the next update, I'm focusing on:
Better prompt templates
Faster testing workflows
More creator-focused examples
If you're interested in AI video workflows, I'm building and testing everything publicly.