Video Generation¶
Two tools cover all video use cases: text-to-video (sana_video_generate) and image-to-video (sana_image_to_video).
Text-to-Video¶
from strands_sana import sana_video_generate
sana_video_generate(
prompt="a slow cinematic pan over snowy mountains, golden hour",
model="sana-video-2b-480", # 480p · 5s clips
steps=15,
frames=121, # ~5s @ 24fps
seed=42,
)
Models for video¶
| Model | Resolution | Default frames | Notes |
|---|---|---|---|
sana-video-2b-480 |
480 | 121 (5s) | Default — best speed/quality |
sana-video-2b-720 |
720 | 121 (5s) | 720p with LTX-VAE refiner |
longsana-2b-480 |
480 | 720 (~27s @ 27fps) | Real-time minute-long |
Output formats¶
sana_video_generate writes:
- MP4 if you have
imageio-ffmpeg(pip install 'strands-sana[video]') - Animated WebP as fallback (no extra deps)
result = sana_video_generate(prompt="...", frames=49)
# {
# "path": "./generated/sana_video_<ts>.mp4",
# "frames": 49, "fps": 24, "duration_s": 2.04,
# }
Image-to-Video¶
Animate a still:
from strands_sana import sana_image_to_video
sana_image_to_video(
image_path="hero.png",
prompt="zoom out slowly, leaves swirling in the wind",
model="sana-video-i2v-480",
frames=49,
seed=42,
)
The source image is encoded into the first frame's latent, then the diffusion model unrolls subsequent frames.
Live demo¶
Resolution constraints¶
SANA-Video requires (height, width) divisible by 32. The wrapper snaps automatically and disables binning to ensure your sizes are honored.
sana_video_generate(prompt="...", height=480, width=480) # OK
sana_video_generate(prompt="...", height=300) # → snaps to 288
Long videos¶
For minute-long real-time clips: