Image to Video¶
Generate an MP4 conditioned on a single first frame. Typical use cases: add camera motion to product shots, animate illustrations.
Minimal example¶
Key parameters¶
| Parameter | Type | Default | Notes |
|---|---|---|---|
image |
str |
required | first-frame path |
prompt |
str? |
None |
motion hint (supported by Wan2.2 i2v and LTX2) |
num_frames |
int? |
model default | SVD typically 14 or 25; Wan2.2 81 |
fps |
int? |
model default | output frame rate |
motion_bucket_id / frame_bucket |
int? |
model default | SVD-only: motion intensity |
noise_aug_strength |
float? |
model default | SVD-only: input-noise perturbation |
decode_chunk_size |
int? |
model default | lower when VRAM is tight |
preset |
fast/balanced/quality/low-vram |
balanced |
see Presets |
seed |
int? |
random | reproducibility |
Supported models¶
- Stable Video Diffusion family:
svd(14 frames),svd-xt(25 frames) - Wan2.2 i2v:
wan2.2-i2v-14b - LTX-Video 2 i2v:
ltx2-i2v
Full list: omnirt models --task image2video.
Common recipes¶
- SVD short clip:
model=svd-xt, num_frames=25, fps=7, motion_bucket_id=127 - Wan2.2 long shot:
model=wan2.2-i2v-14b, num_frames=81, fps=16, prompt="camera panning left"
Troubleshooting¶
Warning
- First frame not aligned — SVD requires
1024×576or576×1024; Wan2.2 i2v resizes to supported buckets automatically - Too much / too little motion — tune
motion_bucket_id(SVD) or describe motion in theprompt(Wan2.2) - Decode OOM — lower
decode_chunk_size(e.g. SVD default →4)