Key Take Aways:
- Exhausitive detail is critical in video creation
- Using a form (my idea) helps organize thoughts and generate consistent results
- Video generation is stateless – need to do each prompt from scratch (form helps having a static part and a changing part)
- Video generators have different sensativities
- Use AI to learn about how to use AI
I’m continuing my quest to learn how to create consistent videos with AI. This is so critical to all aspects of production.
Very important information that I hinted at previously, but Gemini confirmed:
You cannot load two separate documents (a “master” and a “scene-specific” one). AI video generation models currently work from a single text input for each generation. They are “stateless,” meaning they do not remember your previous prompts or have a memory of a “master file” you are referencing.
Think of the AI as a hyper-specific chef. For every single dish you order (each video clip), you must give them the entire recipe from scratch. You can’t just say “make it like the last one, but add salt.” You have to give them the full ingredient list and instructions again, with the addition of “add salt.”
Another critical bit of discovery:
The most important thing to remember is that while this Google Sheet format is perfect for you, the AI still needs one continuous block of text.
I ran into a content flag because of the subject matter. Gemini “cleaned up” my prompt and allowed me to create this!
I then used ChatGPT to give me some facts about the battle which I used to create the dialog for Scene 2 – to give some learning info.
In a first for me, I was able to create two separate scenes that actually match!!
Still some tweaking to be done. I asked why it ignored some of the details like fallen and marching soldiers in the background and it said that AI will do that sometimes to “conserve energy” lol
Here’s a link to my prompts and more info. Feels like a real breakthrough for me
https://duke.box.com/s/q9qqknj8tde555i8dbkdmicuwmdi6z3k
