What about using some stock footage API instead of generating images? I haven't tried it myself, but it should be more cost-effective since the cost should be per month and not per token. This way, you can ask GPT API to generate the search terms for the correct stock footage category and select a random stock video using their API. (I guess this is more or less what InVideo AI does, and I wanted to build a tool like that myself)
Very Insightful and interesting as usual. Love it! Just wondering if Blender's 3D models and rigs can be incorporated into this workflow. It would be intriguing to see how they can be manipulated within the context of this video creation process.
Destroy??! Heck NO! a DIAMOND! 💎 to be polished more is what it is. I saw you got bored with it? about 3/4 of the way. We're learning a Lot! Please complete the polish! Make this Diamond 💎 glisten!!
Use a text to video model, or with those images use a image to video model, use youtube api so it auto uploads the shorts, and use web scraping for finding articles, it could post about 1,000 shorts per day, with that much content it is more probable to get a popular one by luck, your lacking vision, keep improving this project
I'm feelin'..Ya! Hope he keeps it up though. Cause it's just a matter of time... Until someone else gets it right! So keep saving your Bread and Cookies!
@@unconv Yes you are right, I made longer movies of 3 minutes, and TTS1 is not that expensive, nor is GPT3.5, it's the Dall-e that quite dear. I have a system with an Nvidia, I want to try to run it from Stable Diffusion locally, I don't know if I can use something like LiteLLm or ollma to use the openai protocol locally. With solar and SD locally you would have only the TTS to pay for, unless there is an opensource TTS model. But potentially, with the current latest MOE models you could potentially run it for free.