Spaces:
Paused
Paused
| title: "TikSlop" | |
| emoji: 🍿 | |
| colorFrom: red | |
| colorTo: red | |
| sdk: docker | |
| app_file: api.py | |
| pinned: true | |
| short_description: The 100% Latent Video App | |
| models: | |
| - microsoft/Phi-3-mini-128k-instruct | |
| - Lightricks/LTX-Video | |
| app_port: 8080 | |
| disable_embedding: false | |
| hf_oauth: true | |
| hf_oauth_expiration_minutes: 43200 | |
| hf_oauth_scopes: | |
| - inference-api | |
| # TikSlop | |
| ## Configuration | |
| ### WebSocket Connection | |
| - **Web Platform**: Automatically connects to the host serving the page (adapts to both HTTP/HTTPS) | |
| - **Native Platforms**: | |
| - Production: Uses `wss://tikslop.com/ws` when built with `--dart-define=PRODUCTION_MODE=true` | |
| - Development: Uses `ws://localhost:8080/ws` by default | |
| - Custom: Set `API_WS_URL` during build with `--dart-define=API_WS_URL=ws://your-server:port/ws` (highest priority) | |
| ## News | |
| #tikslop is coming sooner than expected! | |
| Stay hooked at @flngr on X! | |
| ## What is TikSlop? | |
| #tikslop is a reboot of [TikSlop 1](https://x.com/danielpikl/status/1737882643625078835), a project made in 2023 which generated AI videos in the background using LLM agents, to simulate an AI generated video platform. | |
| In [#tikslop](https://x.com/flngr/status/1864127796945011016), this concept is put upside down: now the content is generated on demand (when the user types something in the latent search input) and on the fly (video chunks are generated within a few seconds and streamed continuously). | |
| This allows for new ways of consuming AI generated content, such as collaborative and interactive prompting. | |
| # Where can I use it? | |
| #tikslop is not ready yet: this is an experimental side project and the [platform](https://tikslop.com), code and documentation will be in development for most of 2025. | |
| # Why can't I use it? | |
| As this is a personal project I only have limited ressources to develop it on the side, but there are also technological bottlenecks. | |
| Right now it is not economically viable to operate a platform like TikSlop, it requires hardware that is too expensive and/or not powerful enough to give an enjoyable and reactive streaming experience. | |
| I am evaluating various options to make it available sooner for people with the financial ressources to try it, such as creating a system to deploy render nodes to Hugging Face, GPU-on-demand blockchains.. etc. | |
| # When can I use it? | |
| I estimate it will take up to 1 to 2 years for more powerful and/or cheaper hardware to become available. | |
| I already have a open-source prototype of TikSlop which I use for R&D, based on free (as in "can run on your own machine") AI video models that can run fast with low quality settings (such as LTX Video). | |
| It's not representative of the final experience, but that's a start and I use that as a basis to imagine the experiences of the future (collaborative generation, broadcasted live streams, interactive gaming, and artistic experiences that are hybrid between video and gaming). | |