Sora Revival

Unofficial Sora-inspired video generation powered by Open-Sora v1.2

This Space runs a real Open-Sora inference pipeline. It is not an OpenAI product and does not use the Sora 2 model.

Available now: text-to-video, reference-image-conditioned video, prompt-guided camera wording.
Not available in this backend: driving-video motion transfer, pose/skeleton control, or deterministic camera-path control.

Generation mode

Text-to-video Image-to-video

Prompt

Reference image (used only for Image-to-video)

Prompt-guided camera direction

This adds natural-language guidance to the prompt; it is not a dedicated camera-control model.

Resolution

Aspect ratio

Duration

Sampling steps

10 40

Seed

Generated video

Motion control status

Reference-video motion transfer: unsupported by this Open-Sora v1.2 pipeline.
Pose/skeleton guidance: unsupported by this pipeline.
Camera movement: prompt guidance only; not guaranteed or independently controlled.

Do not upload a driving video expecting its motion to be transferred. These controls require a different model or a separately integrated conditioning pipeline.

Runtime note: Open-Sora is a large model. ZeroGPU time and memory are limited; start with 240p / 4 seconds. If the model exceeds the allocation budget or runs out of memory, use a larger dedicated GPU Space. The first run also downloads model weights.