README.md

metadata

title: Wan 2.2 I2V 14B · Custom lora + CivitAI
emoji: 🎬
colorFrom: indigo
colorTo: pink
sdk: gradio
sdk_version: 6.24.0
app_file: app.py
pinned: true
short_description: Image to video, your lora + CivitAI, stitching, cost readout
tags:
  - video
  - image-to-video
  - wan
  - wan2.2
  - lora
  - civitai
  - zerogpu

Wan 2.2 I2V 14B · Custom lora

Turn a still image into smooth cinematic motion, with your own adapters and without guessing what a run will cost.

What is in here

Your own lora, from anywhere. Five slots that take a Hugging Face repo, a file inside one, or a CivitAI download link pasted straight into the High or Low box. Wan adapters ship their two halves as separate files, so each slot has one box for each. Downloads are cached, and a gated CivitAI model works once CIVITAI_TOKEN is set under Settings → Variables and secrets — the Space downloads on its own machine, so a browser session on CivitAI does not authenticate it.

Built-in lora. Three preset slots drawing on a curated catalogue, each with its own high and low strength.

Fast options. Speed presets that set the steps, the flow shift and the scheduler in one click, including the lightx2v Lightning adapters at 4, 6 and 8 steps. The checkpoint is a Lightning merge already, so Built-in is the place to start.

ZeroGPU cost, before you press the button. The readout above Generate uses the very formula the Space reserves with, so heavy settings are visible as a number rather than as a failed run.

Clip stitching and looping, on the CPU. Queue several clips and they are joined into one file; tick the box and every new generation joins on its own. Boomerang loops the whole clip or just its tail. None of it touches the GPU allowance.

Named profiles. Save a whole set-up under a name and bring it back from a dropdown, or export it as .json.

Frame grabbing. Pause the result on any frame and send it to the start image or the end image, so a scene can continue from where it stopped.

Notes

• fp8 quantized, AoT compiled, with RIFE interpolation for smoother motion.
• Longer clips reserve more ZeroGPU time; the cost line tells you how much.
• Profiles live inside the Space. With persistent storage attached they survive a restart, and they are visible to everyone who opens the Space.

Configuration reference: https://huggingface.co/docs/hub/spaces-config-reference