メニュー
Generate videos with vLLM-Omni using Wan2.2 and other video generation models. Use when generating videos from text, creating videos from images, configuring video generation parameters, or working with text-to-video or image-to-video models.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Review PRs on vllm-project/vllm-omni by routing to the right domain skills, checking critical evidence, and focusing comments on blocking issues. Use when reviewing pull requests or local branches, triaging review depth, running detailed or default review, or checking tests, benchmarks, and breaking changes in vllm-omni.
Two report kinds; **default output is always HTML** unless the user explicitly asks for Markdown (.md). **Release** — `scripts/compose_full_report.py` (**测试结论**, Buildkite metrics, **Test Result** = Common stack + optional `--log-dir-h*` nightly-style summaries + H100/CI block, **Issue tracking** = GitHub `ci-failure` + *local test* in:title, Open bugs); use `--format markdown` only when the user wants .md or `patch_report_*.py`. **Nightly** — `scripts/nightly_local_log_report.py` from local `nightly_jobs` (fetch: vllm-omni-nightly-local) plus optional latest Buildkite scheduled nightly when token is set; use `--markdown-report` / `--to-stdout markdown` only when the user asks for Markdown. Use when generating Buildkite release summaries, parsing local nightly_jobs with CI cross-check, or opening https://buildkite.com/vllm/vllm-omni/builds?branch=main for CI documentation.
Use before submitting a PR to vllm-project/vllm-omni — self-check the branch against project conventions, catch dead code, verify accuracy/performance claims, and confirm merge readiness. Use when the user says "pre-check", "self review", "pre-submit check", or "check my PR before I open it."
Generate audio and speech with vLLM-Omni using Qwen3-TTS, Fish Speech S2 Pro, CosyVoice3, MiMo-Audio, and Stable-Audio models. Use when synthesizing speech from text, generating audio effects or music, configuring TTS parameters, cloning voices, adding new TTS models, or working with text-to-speech models.
Contribute to vLLM-Omni by adding new model support, fixing bugs, or improving features. Use when integrating a new model into vllm-omni, setting up a development environment, writing tests, or submitting pull requests to the vllm-omni project.
Scale vLLM-Omni across multiple GPUs and nodes using tensor parallelism, pipeline parallelism, OmniConnector disaggregation, connector backends, and Ray. Use when setting up multi-GPU inference, distributing model execution across machines, deploying disaggregated execution, developing OmniConnector backends, or scaling inference horizontally.
| name | vllm-omni-video-gen |
| description | Generate videos with vLLM-Omni using Wan2.2 and other video generation models. Use when generating videos from text, creating videos from images, configuring video generation parameters, or working with text-to-video or image-to-video models. |
vLLM-Omni supports video generation through diffusion transformer models, primarily the Wan2.2 family. Three modes are supported: text-to-video (T2V), image-to-video (I2V), and text+image-to-video (TI2V).
| Model | HF ID | Mode | Min VRAM |
|---|---|---|---|
| Wan2.2-T2V-A14B | Wan-AI/Wan2.2-T2V-A14B-Diffusers | Text-to-video | 48 GB |
| Wan2.2-TI2V-5B | Wan-AI/Wan2.2-TI2V-5B-Diffusers | Text+Image-to-video | 24 GB |
| Wan2.2-I2V-A14B | Wan-AI/Wan2.2-I2V-A14B-Diffusers | Image-to-video | 48 GB |
| NextStep-1.1 | stepfun-ai/NextStep-1.1 | Text-to-video | 24 GB |
| Helios-Distilled | naver-ai/Helios-Distilled | Text-to-video | 24 GB |
| daVinci-MagiHuman | SII-GAIR/daVinci-MagiHuman-Base-1080p | Image-to-video + audio | 24 GB |
daVinci-MagiHuman is an image-to-video model that also generates audio (44100 Hz, 25 fps). Use --enable-diffusion-pipeline-profiler to get per-stage timing (stage_durations) and peak memory (peak_memory_mb) in video responses (async poll JSON or sync HTTP headers).
from vllm_omni.entrypoints.omni import Omni
omni = Omni(model="Wan-AI/Wan2.2-T2V-A14B-Diffusers")
outputs = omni.generate("A dog running on a beach at sunset")
video = outputs[0].request_output[0].video
video.save("dog_beach.mp4")
vllm serve Wan-AI/Wan2.2-T2V-A14B-Diffusers --omni --port 8091
curl -s http://localhost:8091/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"messages": [{"role": "user", "content": "A dog running on a beach at sunset"}],
"extra_body": {
"num_inference_steps": 50,
"guidance_scale": 5.0,
"seed": 42
}
}'
Animate a static image into a video:
from vllm_omni.entrypoints.omni import Omni
omni = Omni(model="Wan-AI/Wan2.2-I2V-A14B-Diffusers")
outputs = omni.generate(
prompt="The person starts walking forward",
images=["portrait.jpg"],
)
outputs[0].request_output[0].video.save("animated.mp4")
Combine a text description and reference image:
omni = Omni(model="Wan-AI/Wan2.2-TI2V-5B-Diffusers")
outputs = omni.generate(
prompt="The city lights up at night with moving traffic",
images=["cityscape.jpg"],
)
outputs[0].request_output[0].video.save("city_night.mp4")
| Parameter | Description | Typical Range |
|---|---|---|
num_inference_steps | Denoising steps | 30-100 |
guidance_scale | CFG scale | 3.0-7.0 |
seed | Random seed | Any integer |
num_frames | Number of output frames | Model-dependent |
fps | Frames per second | 8-24 |
Video generation is significantly more compute-intensive than image generation:
vllm serve <model> --omni --cpu-offload-gb 20
transformer + transformer-2), the sequential offloader now offloads all other DiTs to CPU when any one is running. This allows Wan2.2-T2V to fit on 64GB GPUs with --enable-cpu-offload --tensor-parallel-size 2.Generation too slow: Use tensor parallelism or enable TeaCache/Cache-DiT acceleration. Helios supports cache-dit (--enable-cache-dit) for ~20% speedup.
LTX-2 error with diffusers>=0.38.0: Fixed in #3661. Text encoder normalization moved into the diffusers connector. Update vllm-omni to the latest version when upgrading diffusers to 0.38.0+.
Out of memory: Reduce resolution/frame count or use CPU offloading.
Choppy output: Increase num_inference_steps and num_frames.