Hello, friendly people!
I wanted to share the workflow behind Chris & Saleem Save the World, including a few tricks that helped me keep generation costs under control. I’ve seen a lot of projects here come together across several platforms and editing tools. For this one, I wanted to bring the whole process into a single app and make it simple and affordable—because generation costs can get out of hand quickly.
I built Zapters alongside the project, shaping the app around the workflow as I went. Some of this will probably be familiar, but there are a few techniques I haven’t seen discussed much that I thought would be worth sharing.
Project setup and Characters
I wanted one place where I could see and organize everything: episodes (called “Zapters” in the app), characters, locations, storyboards, and music.
For characters, the idea was simple: describe someone and get a consistently structured character sheet, without having to choose a model every time. The default is currently GPT Image 2.5, which I’ve found offers a good balance of quality and cost, though you can change it in the settings. I also wanted each character to have their own voice. You can either design one with ElevenLabs or clone a voice from a generated video.
Scripting
I built the scripting system around beats, each up to 15 seconds long. A beat contains individual shots with timecodes, such as 00:00–00:05, plus camera settings for movement, angle, and framing. Those settings are easy to swap, which makes experimenting with different approaches much quicker.
The screenshots below show the scripting screen. Since the look of a scene matters a lot to me, I added lighting controls too, including time of day and light temperature. What I like about this approach is that it feels more like planning a scene than writing a prompt. The app turns those choices into a prompt tailored to the generation model.
Each beat can also have a start frame, an end frame, and its own storyboard. I can generate a storyboard directly from the beat with a tap, which makes it easier to plan the visuals before moving into video.
Video generation (Directing)
This is where most of my cost-saving tricks come in. After learning the hard way that AI rarely gets everything right on the first attempt, I started thinking of generations as takes. Each beat can have multiple takes. In the screenshots, the indicators on the right show those takes, while the top shows how many beats have generated video.
Creating a new take is as simple as tapping the beat I want to generate. The app automatically selects a take from the previous beat to continue from. Referencing the previous video helps maintain consistency, but it also increases the cost. To reduce that, I built “Smart Take,” shown by the light bulb icon. It turns the previous video into a filmstrip image that can be used as a visual reference.
Another useful trick is to try cheaper models once the visual direction is established. Seedance 2.0 Mini has worked well for many of my scenes, especially when I use a scene generated with Seedance 2.0 as the visual anchor. I often start with Mini at 480p to check whether the scene, pacing, and timing work. If they don’t, the failed experiment costs much less. Sometimes I like the draft enough to keep it and use the built-in Topaz upscaling to bring it to 720p.
Sometimes a beat contains several good shots and one difficult shot that lets it down. If I already have multiple takes, I can combine the best shots directly in the app. Otherwise, I can select just the shots I want to regenerate—using Seedance 2.0 for a more demanding shot, for example—without paying to redo the entire beat.
Once I have a take I’m happy with, I use the voice changer to replace the generated voices with the ones I designed for my characters. Applying the voices at this stage keeps costs down and works around the limitations of using audio references in Seedance 2.0.
Combining (Editing)
Once the takes are ready, I move into editing, which also happens inside the app. Auto Combine assembles the beats as they’re generated, so I can watch the sequence come together. For finer adjustments, I can open the timeline editor. For this project, I added music using Lyria 3, rendered the edit, and enabled the built-in captions. That brought the whole process—from character creation to the finished video—together in one app.
Final word
Working on Chris & Saleem has been a blast. It brought together two things I love: telling stories and building software. Developing the app alongside an actual production helped me figure out which features were useful in practice, and the feedback from users has been encouraging. I’d love to hear how you approach your own projects. Do you use any of these techniques? And what workflow tips would you recommend I try next?
If you are intrested you can checkout Zapters here: zapters.app