- Joined
- Mar 13, 2023
- Messages
- 460
- Reaction score
- 473
- Points
- 63
It uses the LLAMA2 Instruct Model for the content, the Bark TTS Model for Voice, and Videos from Pexels. No OpenAI API Key, no expensive GPU servers, everything is free. All you need is a pexels API Key. Make Free Videos with Near Studio Quality Human-like audio
The LLM and the TTS models run on the GPU, but the video maker kind of runs on the CPU and you don't get the best CPUs on colab and you migth be better off running that part on your local machine. Should be easy if you know how to code, otherwise hard luck.
Steps to run:
The problem with Run All is that the LLM may generate stupid text in the output of some prompts, and you might want to edit that. While the Run All should be good for the initial test when you really want to make some good stuff you might want to run one cell at a time. Once the Content is generated you will need to go to the drive and pull the json file that is created and review the content and make edits as needed and upload it back to your drive without messing up the format and then start with the Audio section. You may also want to play with the prompts. The text generation structure is explained in the notebook.
The Audio Model has a number of voice presets, that can be found at
I personally like the v2/en_speaker_6 and v2/en_speaker_7 as they are nearly studio quality voices while the others kind of sound like Alexander Graham Bell's grandfather on the phone.
The Video making is quite slow on Colab as they don't give you the best CPUs, but it does the job over time. If you don't know how to setup your local environments and run the code locally, it might be best to keep it in colab itself.
Happy to offer support but all questions need to go on this thread. I will NOT respond to stupid questions. If you try to reach me in any other way, I will track you down and hand you over to Mossad and tell them you're Hamas.
Disclaimer: The LLAMA2 and the Bark Models come with their own license terms, and it is the sole responsibility of the users to adhere to them and by using this code, you understand and agree that I shall in no way be held responsible for any oversights or breaches of their licenses if Zucky decides to come after you.
Warning: The code and the docs cuss a lot.
Enjoy!
The LLM and the TTS models run on the GPU, but the video maker kind of runs on the CPU and you don't get the best CPUs on colab and you migth be better off running that part on your local machine. Should be easy if you know how to code, otherwise hard luck.
Steps to run:
- Click on the link above and save the file to your drive.
- Click on Runtime and Change the Runtime type to GPU
- Enter a topic within the quotes of the t0pic Variable. If you need Travel Videos about Rio De Janeiro, just type Rio De Janeiro.
- Add your pexels key in the variable below that.
- Click on runtime again and hit Run All
- Allow google drive permissions where it saves your tmp files and the output.
- Come back in an hour and it should be done by then.
The problem with Run All is that the LLM may generate stupid text in the output of some prompts, and you might want to edit that. While the Run All should be good for the initial test when you really want to make some good stuff you might want to run one cell at a time. Once the Content is generated you will need to go to the drive and pull the json file that is created and review the content and make edits as needed and upload it back to your drive without messing up the format and then start with the Audio section. You may also want to play with the prompts. The text generation structure is explained in the notebook.
The Audio Model has a number of voice presets, that can be found at
I personally like the v2/en_speaker_6 and v2/en_speaker_7 as they are nearly studio quality voices while the others kind of sound like Alexander Graham Bell's grandfather on the phone.
The Video making is quite slow on Colab as they don't give you the best CPUs, but it does the job over time. If you don't know how to setup your local environments and run the code locally, it might be best to keep it in colab itself.
Happy to offer support but all questions need to go on this thread. I will NOT respond to stupid questions. If you try to reach me in any other way, I will track you down and hand you over to Mossad and tell them you're Hamas.
Disclaimer: The LLAMA2 and the Bark Models come with their own license terms, and it is the sole responsibility of the users to adhere to them and by using this code, you understand and agree that I shall in no way be held responsible for any oversights or breaches of their licenses if Zucky decides to come after you.
Warning: The code and the docs cuss a lot.
Enjoy!
