Models¶

strands-sana registers 15 checkpoints across 7 pipeline kinds.

List at runtime¶

from strands_sana import sana_load_model

print(sana_load_model(model="list"))
print(sana_load_model(model="list", kind="video"))      # only video
print(sana_load_model(model="list", tag="sana-1.5"))    # only Sana-1.5

Image models¶

Alias	Pipeline	Resolution	Params	Use case
`sana-0.6b-512`	t2i	512	590M	Fast iteration
`sana-0.6b-1024`	t2i	1024	590M	Lightweight 1024
`sana-1.6b-1024`	t2i	1024	1.6B	Default
`sana-1.6b-multiling`	t2i	1024	1.6B	Chinese/Emoji
`sana-1.5-1.6b-1024`	t2i	1024	1.6B	Improved GenEval
`sana-1.5-4.8b-1024`	t2i	1024	4.8B	Top quality
`sana-1.6b-2k`	t2i	2048	1.6B	2K resolution
`sana-1.6b-4k`	t2i	4096	1.6B	4K resolution
`sana-sprint-0.6b-1024`	sprint	1024	590M	1-2 step, ultra-fast
`sana-sprint-1.6b-1024`	sprint	1024	1.6B	1-2 step, top quality
`sana-sprint-i2i-1.6b-1024`	sprint-i2i	1024	1.6B	Image-to-image

Video models¶

Alias	Pipeline	Resolution	Frames	Use case
`sana-video-2b-480`	video	480	121	Default 5s clip
`sana-video-2b-720`	video	720	121	720p with LTX-VAE
`longsana-2b-480`	video	480	720	Real-time minute-long
`sana-video-i2v-480`	image-to-video	480	121	Animate a still

Pipeline kinds¶

graph TB
    Wrapper["SanaPipelineWrapper"] --> T2I["t2i → SanaPipeline"]
    Wrapper --> PAG["pag → SanaPAGPipeline"]
    Wrapper --> Sprint["sprint → SanaSprintPipeline"]
    Wrapper --> SprintI2I["sprint-i2i → SanaSprintImg2ImgPipeline"]
    Wrapper --> Ctrl["controlnet → SanaControlNetPipeline"]
    Wrapper --> Vid["video → SanaVideoPipeline"]
    Wrapper --> I2V["image-to-video → SanaImageToVideoPipeline"]

Per-model defaults¶

Each model has tuned default_steps, default_guidance, default_frames. The wrapper uses these unless you override:

from strands_sana.models import get_model_info

info = get_model_info("sana-sprint-1.6b-1024")
print(info.default_steps)     # 2 — Sprint sweet spot
print(info.default_guidance)  # 0.0 — distilled, no CFG

Pre-download a checkpoint¶

from strands_sana import sana_prefetch_model

sana_prefetch_model(model="sana-1.6b-1024")
# Returns: { "local_path": "~/.cache/huggingface/hub/..." }