Wan2.1 T2V 1.3B — video from text, for xOllama

Wan2.1 T2V 1.3B · umt5-xxl text encoder · Wan2.1 VAE · 640×352, 33 frames at 16 fps · OpenAI /v1/videos

Runs on xOllama, not on stock ollama. Stock ollama pulls this model but cannot show or run it: it carries only media engines. xOllama serves it through an opencoti engine with video support and its codec sidecar (dev build 2610022033001/b105 or later, set with XOLLAMA_ENGINE_PATH). The engine pinned by the current xOllama release has no media engines and refuses the model at load, naming what it lacks.

Short clips from a text prompt, in the least memory of the video models here.

Use

A clip is a job, as in OpenAI’s videos API: create it, poll it, fetch it.

xollama pull mannix/wan2.1:t2v-1.3b
curl http://localhost:22434/v1/videos \
  -d '{"model": "mannix/wan2.1:t2v-1.3b", "prompt": "a red fox running through fresh snow at sunrise"}'
# {"id": "video_…", "object": "video", "status": "queued", …}
curl http://localhost:22434/v1/videos/video_…            # until "status": "completed"
curl http://localhost:22434/v1/videos/video_…/content -o fox.mp4

The template’s clip

A request with no size gets 640×352, 33 frames at 16 fps (2 s), cfg 6, euler, flow shift 3, h264 mp4. That is also the largest clip the template serves: its 12 GiB working reserve is sized for it, so a 24 GB card holds it. A bigger request is refused with a 400 before anything loads. The same pixel count in another shape (352×640), or a smaller size with more frames, is fine.

Measured

At 832×480 the VAE decode alone wants about 19 GB, so on a 24 GB card the engine tiles it and the clip takes 563–570 s. That is why the template’s clip is the smaller one.

Tags

t2v-1.3b is the one for a 24 GB card. t2v-1.3b-fp16 keeps full precision and peaks 3.4 GiB higher.

Source

samuelchristlie/Wan2.1-T2V-1.3B-GGUF · fp16 weights and VAE: Comfy-Org/Wan_2.1_ComfyUI_repackaged · text encoder: city96/umt5-xxl-encoder-gguf · License: Apache-2.0. Docs: xOllama media, video jobs.