v0.2.0-alpha

Vikit Reloaded

V0.2 : Faster, more providers and Image Prompts
Now you have the possibility to use Image Based Prompt, kinda cool! 😎 Why doing this? You may want to start your new scenes from previous scenes context (for instance leverage get_last_frame_as_image), or guide the video generation models with a given aesthetic.

Vikit 0.2 now implements async (at last 😉) and it is combined with multi processing support, which means video should take an average of 5 to 7 minutes to be fully generated, whatever the size of the prompts, depending on your own hardware, i.e. network bandwidth and CPU for ffmpeg video editing.
Async allows for gathering the different video scenes without blocking the main video building workflow, as if it where all done in parallel.
Multiprocessing is used to leverage all your computer cores when editing the video with all those generated or imported videos, as well as background music, read aloud prompt, or re-encoding.
Warning: Too many parallel video generation processes might slow things down, just keep an eye on your local CPU usage.
Note: a Dev Environment version is in the works to get you started quickly , so you may benefit from Github x core VM's if needed. Also we received feedback that code is cool though a PIP version would be nice too for direct use.

Last but not least, we now integrate with Stability AI (for testing purpose) and Haiper, and still have the possibility to use VideoCrafter! The first two providers let you leap forward in the quality of generated video, enjoy! Important note: Haiper do work, though we are working with their teams to figure out an operational issue , be sure we do everything for this feature to be back soon 🙂

Stay tuned for new integrations to come on the Video generation and music generation space

Some other features you might want to experiment:

Prepare a video (prepare_build) with specific build settings, which means choosing your video generation provider, music, prompt, etc...all that can be applied with VideoBuildSettings object

Also, the video generation process is streamlined trough two ways:

You can now use Hooks to get your code run adjust before the build (run_pre_build_actions_hook) , during (run_build_core_logic_hook) and after build (run_post_build_actions_hook). Not too hard, huh?

Handlers! just stack handlers in the order you want and let the process happen! You shall adapt or implement get_core_handlers function in the different type of Videos available to date set your own handlers if needs be...

Two additional features we hope you will find handy:

It is now possible to specify a target path where to save your videos. It could be a remote path provided you access it without authentication. Feel free to submit pull requests if you strongly need support for Cloud buckets or other target file systems.

You may cascade build settings, which means all the tree made of your composite video will share the same overall settings

Happy coding! 🚀😊

v0.1.0-alpha

0.1.0-alpha - Codename: Frame

This is the very first version, experimental MVP, of Vikit.ai SDK.

It was originally private source and allows generating videos from an audio recording (e.g. a blog audio) or text. It also allows for merging the final video with background music and read aloud the prompt generated from text.

Music can be generated, or a default royaltee free background music. You may also use your own prompt recording as an audio track.

0.1 uses various models for video generation, music generation, audio prompt to text and synthetic voice generation.

Available type ov videos are:

• RawTextBasedVideo: the most basic building block that usually corresponds to a 2 to 6s video , depending on the generative model and platform providing it
• CompositeVideo: A video made of other video. You will probably play a lot with this beast and the RawTextbasedVideo!
• PromptBasedVideo: for lazy people like us 😉 who just want to give a prompt and let the video be generated. it inherits composite video and includes a first simple way to stitch video and transitions together
• Imported (i.e. your video, generated or not)

Initial beta testers experimented with 0.1 using a Google Colab, though video could take a long time to generate if you stacked many scenes within (tens of minutes).