Image-to-video

Official commands

python generate.py --task i2v-A14B --size 1280*720 --ckpt_dir ./Wan2.2-I2V-A14B --offload_model True --convert_model_dtype --image examples/i2v_input.JPG --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. …"

python generate.py --task ti2v-5B --size 1280*704 --ckpt_dir ./Wan2.2-TI2V-5B --offload_model True --convert_model_dtype --t5_cpu --image examples/i2v_input.JPG --prompt "…"

If --image is set on ti2v-5B, the run is image-to-video; otherwise it is text-to-video. That is the hybrid design the README advertises.

Image-only (empty prompt)

I2V can run with --prompt '' when --use_prompt_extend is on. DashScope qwen-vl-max (or local Qwen2.5-VL) writes the motion prompt from the still. That is the supported way to “just upload a photo.”

What the start image actually controls

• Subject identity and wardrobe at t = 0.
• Output aspect ratio (letterboxing if you force a mismatched size).
• Lighting and background as a prior — not a locked plate. Large camera moves can drift.

FAQ

480P or 720P?

I2V-A14B supports both. TI2V-5B is the 720P@24 fps consumer hybrid. Pick A14B for quality if you have 80 GB-class GPUs.

Does I2V keep the face still?

It tries. For a locked character plus motion from another clip, use Animate-14B or VACE reference — not vanilla I2V.