Spaces:

---

FrameAI4687

/

Omni-Video-Factory

like 871

Running on Zero

Good question — happy to share some best practices from our build process.
We actually ran into quite a few pitfalls while setting this up and learned a lot along the way.

1. Do NOT put ComfyUI directly into the Hugging Face Space repository.
   This is very important. If you embed the full ComfyUI codebase inside the Space repo, the Space may fail to restart properly after being paused or rebuilt.
   To avoid this, we host ComfyUI separately here:
   https://github.com/selfitcamera/ComfyUIVideo
   and download it automatically when the Space starts.

2. Model choice
   Our production model is OmniVideo, a model we trained ourselves.
   That said, our workflows are intentionally designed to be flexible. The workflows in:
   https://github.com/selfitcamera/ComfyUIVideo/tree/main/workflows
   also support open-source models such as Wanx 2.2 / LTX-2, and you can freely swap in any compatible model.

3. Workflow reuse without repeated model loading
   Although we provide three different workflows, they all share the same underlying model.
   This allows us to reuse the loaded model in memory instead of reloading it multiple times, which saves a lot of startup and execution time.

4. Automation with Codex
   The Space logic is mostly implemented here:
   https://huggingface.co/spaces/FrameAI4687/Omni-Video-Factory/blob/main/app.py

   In practice, you can take this file together with the GitHub repository, give them to Codex, configure your Hugging Face and GitHub SSH keys, and Codex can handle most of the setup for you — including adapting the pipeline to any model you want.

Hope this helps explain why we designed it this way and makes it easier for others to build similar systems.

Thanks.

This is my video on this platform. I used this prompt with 1 scene , 3 sec approach, 318 use. With an image shown within the gif video attached link. I do not know now how to provide more accurate results but I am trying to enhance and correct my prompt for the video. Here it is" Try to create a morning freezing boy plane group swimming storymat on groundview of an in play frames drawn voxel water zigzag mat group". I used this image as reference. (https://cdn-uploads.huggingface.co/production/uploads/67e69db89bbc79d446e86427/ZeAiu8kIKZqJvEMkH7UmB.png) which I had previously created with Adobe firefly.