Add a short scene
Start with a continuous 3-second shot. Refer to it as <Video 1>.
Bring a reference character into your video. Add a scene, choose a portrait or character sheet, and describe who to replace.
Start with a continuous 3-second shot. Refer to it as <Video 1>.
Upload a clear portrait or character sheet. This is <Picture 1>.
Name the person to replace, then generate. Or try an example below.
Author-published comparison · V1 on the left, V1.5 on the right
Live demo · shared ZeroGPU backend
For your first run: choose the cafe + hiker example, then try a smaller canvas in Advanced options. Generation can take several minutes. Our minimum-size tests exceeded the free-account GPU duration limit; a PRO account or sufficient quota may be needed to generate. You can watch the published comparison above without using GPU quota. If the embedded app asks you to sign in, use “Open full demo” above.
V1.5 is the model author's recommended adapter. Identity, timing, expressions, occlusions and scene details can still drift. Short continuous shots work best. This demo generates its soundtrack. The example inputs come from the V1 training dataset.
This free Space embeds hugging-apps' live inference demo. Uploaded media and prompts go to that service and its conditioning Space. Availability and quota are managed by those services. The standalone Gradio app is included for hosting on your own eligible ZeroGPU account.