This is a toy project I've been hacking on, and it's a bit rough around the edges, but I'd love for people to try it out! https://aitrailers.xyz
Some examples of videos people have generated:
* An action movie starring a corgi: https://aitrailers.xyz/examples/paws-of-power
* A children's movie about a mischievous monkey: https://aitrailers.xyz/examples/cheeky-monkey
* And a "thriller" my friend Mark generated about how he lost one of his Airpods in a tissue box: https://aitrailers.xyz/examples/marks-lost-earbud
So far it's been fun to see what people have come up with, but the quality of the output varies quite a bit... I'd love to get any feedback on the app so far, and also curious if anyone has any advice on the following:
* What is the best way to improve character consistency across images? (I currently just try to construct each image prompt such that a vivid visual description of the main character and the overall aesthetic is the same across images, but I've had limited success with this.)
* Have any other tools caught up to GPT-4 and DALL-E for things like this? (I've tinkered with some models on Replicate, but nothing is quite as good as OpenAI's APIs as far as I can tell.)
* Are there any good "wrappers" for ffmpeg? I'm currently using a combination of ffmpeg and moviepy, but it surprises me that there isn't an "easier" way to generate videos. (Ffmpeg is super powerful, but it’s API is a bit unintuitive)
Some examples of videos people have generated:
* An action movie starring a corgi: https://aitrailers.xyz/examples/paws-of-power * A children's movie about a mischievous monkey: https://aitrailers.xyz/examples/cheeky-monkey * And a "thriller" my friend Mark generated about how he lost one of his Airpods in a tissue box: https://aitrailers.xyz/examples/marks-lost-earbud
So far it's been fun to see what people have come up with, but the quality of the output varies quite a bit... I'd love to get any feedback on the app so far, and also curious if anyone has any advice on the following:
* What is the best way to improve character consistency across images? (I currently just try to construct each image prompt such that a vivid visual description of the main character and the overall aesthetic is the same across images, but I've had limited success with this.) * Have any other tools caught up to GPT-4 and DALL-E for things like this? (I've tinkered with some models on Replicate, but nothing is quite as good as OpenAI's APIs as far as I can tell.) * Are there any good "wrappers" for ffmpeg? I'm currently using a combination of ffmpeg and moviepy, but it surprises me that there isn't an "easier" way to generate videos. (Ffmpeg is super powerful, but it’s API is a bit unintuitive)