[Model] Add Wan2.2 I2V and TI2V pipeline support by linyueqian · Pull Request #329 · vllm-project/vllm-omni

linyueqian · 2025-12-16T04:47:13Z

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Add support for Wan2.2 Image-to-Video (I2V) and Text-Image-to-Video (TI2V) pipelines.

Test Plan

Test 1: I2V-A14B (Image-to-Video with MoE)
python examples/offline_inference/image_to_video/image_to_video.py \ --model Wan-AI/Wan2.2-I2V-A14B-Diffusers \ --image input.jpg --num_frames 17 --num_inference_steps 10 \ --output i2v_a14b_output.mp4

Test 2: TI2V-5B I2V mode (Image-to-Video with unified model)
python examples/offline_inference/image_to_video/image_to_video.py \ --model Wan-AI/Wan2.2-TI2V-5B-Diffusers \ --image input.jpg --num_frames 17 --num_inference_steps 10 \ --output ti2v_5b_i2v_output.mp4

Test 3: TI2V-5B T2V mode (Text-only, no image)
python examples/offline_inference/text_to_video/text_to_video.py \ --model Wan-AI/Wan2.2-TI2V-5B-Diffusers \ --prompt "A cat walking in the garden" \ --num_frames 17 --num_inference_steps 10 \ --output ti2v_5b_t2v_output.mp4

Test Result

i2v_a14b_output.mp4

ti2v_5b_i2v_output.mp4

ti2v_5b_t2v_output.mp4

Essential Elements of an Effective PR Description Checklist

The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
The test plan, such as providing test command.
The test results, such as pasting the results comparison before and after, or e2e results
(Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
(Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

Signed-off-by: linyueqian <linyueqian@outlook.com>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py

SamitHuang

please update supported model list as well

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_ti2v.py

Signed-off-by: linyueqian <linyueqian@outlook.com>

hsliuustc0106 · 2025-12-17T06:22:27Z

vllm_omni/diffusion/cache/selector.py

@@ -1,8 +1,6 @@
 from typing import Any


why we need to move the imports?

i didnt install the cache dependency at that time. i can revert it.

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py

examples/offline_inference/image_to_video/image_to_video.py

hsliuustc0106 · 2025-12-17T06:25:14Z

docs/models/supported_models.md

 | `QwenImageEditPipeline` | Qwen-Image-Edit | `Qwen/Qwen-Image-Edit` |
 |`ZImagePipeline` | Z-Image | `Tongyi-MAI/Z-Image-Turbo` |
-| `WanPipeline` | Wan2.2 | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` |
+| `Wan22Pipeline` | Wan2.2-T2V | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` |


use WAN22Pipeline for all t2v, i2v and ti2v models?

The three variants have different architectures (I2V-A14B uses CLIP encoder + MoE, TI2V-5B uses single transformer with VAE conditioning). Consolidation is possible but adds some model selection complexity in one pipeline. I'm leaning toward keeping separate pipelines, which follows diffusers' convention where T2V and I2V also have different pipeline classes. Let me know if you prefer otherwise.

@ZJY0516 WDYT

For the pipeline name, i think we should follow the identifier in Wan2.2-T2V-A14B-Diffusers/
model_index.json

"_class_name": "WanPipeline"

hsliuustc0106

let's get it done asap

Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com> Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>

Signed-off-by: linyueqian <linyueqian@outlook.com>

…o feat/wan2.2-i2v

Signed-off-by: linyueqian <linyueqian@outlook.com>

linyueqian · 2025-12-22T07:25:38Z

@SamitHuang @hsliuustc0106 Please review again. I use the identifier name from their corresponding model_index.json file.

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py

Signed-off-by: linyueqian <linyueqian@outlook.com>

hsliuustc0106 · 2025-12-24T02:32:21Z

could you please submit a relevant recpie to vllm/recipe?

ZJY0516 · 2025-12-24T02:37:13Z

could you please submit a relevant recpie to vllm/recipe?

Currently, we only support Cache-DiT acceleration for wan, so a recipe briefly mentioning this would be perfect.

linyueqian · 2025-12-24T06:17:30Z

could you please submit a relevant recpie to vllm/recipe?

Currently, we only support Cache-DiT acceleration for wan, so a recipe briefly mentioning this would be perfect.

vllm-project/recipes#184

hsliuustc0106

lgtm, later we are going to test USP&CFG parallel when the feature is ready.

Signed-off-by: linyueqian <linyueqian@outlook.com> Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com> Signed-off-by: wangyu31577 <wangyu31577@hundsun.com>

Signed-off-by: linyueqian <linyueqian@outlook.com> Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

Add Wan2.2 I2V and TI2V pipeline support

5947697

Signed-off-by: linyueqian <linyueqian@outlook.com>

linyueqian requested a review from hsliuustc0106 as a code owner December 16, 2025 04:47

Merge branch 'vllm-project:main' into feat/wan2.2-i2v

4d59425

chatgpt-codex-connector bot reviewed Dec 16, 2025

View reviewed changes

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py Outdated Show resolved Hide resolved

hsliuustc0106 requested review from Isotr0py, SamitHuang and ZJY0516 December 16, 2025 04:58

david6666666 mentioned this pull request Dec 16, 2025

[RFC]: DiT model and feature support enhancement #85

Closed

58 tasks

ZJY0516 linked an issue Dec 16, 2025 that may be closed by this pull request

[New Model]: Wan-AI/Wan2.2-I2V-A14B-Diffusers #325

Closed

1 task

SamitHuang reviewed Dec 17, 2025

View reviewed changes

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_ti2v.py Outdated Show resolved Hide resolved

linyueqian added 3 commits December 16, 2025 21:34

update supported model

07ca8ca

Signed-off-by: linyueqian <linyueqian@outlook.com>

avoid duplicate function

4294bd3

Signed-off-by: linyueqian <linyueqian@outlook.com>

loading transformer_2 correctly

6ed4cdf

Signed-off-by: linyueqian <linyueqian@outlook.com>

hsliuustc0106 reviewed Dec 17, 2025

View reviewed changes

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py Show resolved Hide resolved

hsliuustc0106 reviewed Dec 17, 2025

View reviewed changes

examples/offline_inference/image_to_video/image_to_video.py Outdated Show resolved Hide resolved

hsliuustc0106 reviewed Dec 17, 2025

View reviewed changes

linyueqian and others added 4 commits December 17, 2025 15:26

Update examples/offline_inference/image_to_video/image_to_video.py

f52cc03

Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com> Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>

revert changes

491d93f

Signed-off-by: linyueqian <linyueqian@outlook.com>

Merge branch 'feat/wan2.2-i2v' of github.com:linyueqian/vllm-omni int…

26480c2

…o feat/wan2.2-i2v

fix naming

26c9ed9

Signed-off-by: linyueqian <linyueqian@outlook.com>

hsliuustc0106 added the ready label to trigger buildkite CI label Dec 22, 2025

taoyun951753 reviewed Dec 23, 2025

View reviewed changes

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py Outdated Show resolved Hide resolved

taoyun951753 reviewed Dec 23, 2025

View reviewed changes

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2_i2v.py Outdated Show resolved Hide resolved

fix: improve image encoder handling in Wan2.2 I2V pipeline

abfd519

Signed-off-by: linyueqian <linyueqian@outlook.com>

Merge branch 'vllm-project:main' into feat/wan2.2-i2v

cfdf499

hsliuustc0106 reviewed Dec 25, 2025

View reviewed changes

hsliuustc0106 approved these changes Dec 25, 2025

View reviewed changes

hsliuustc0106 merged commit 3ec35e5 into vllm-project:main Dec 25, 2025
7 checks passed

david6666666 mentioned this pull request Jan 9, 2026

[Feature]: vLLM-Omni model owner JiusiServe/vllm-omni#25

Open

17 tasks

david6666666 mentioned this pull request Jan 16, 2026

vLLM-Omni Model Support #808

Open

55 tasks

Comments

Conversation

linyueqian commented Dec 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Test Plan

Test Result

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

SamitHuang left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

hsliuustc0106 Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

linyueqian Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

hsliuustc0106 Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

linyueqian Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

hsliuustc0106 Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

SamitHuang Dec 20, 2025

Choose a reason for hiding this comment

Uh oh!

hsliuustc0106 left a comment

Choose a reason for hiding this comment

Uh oh!

linyueqian commented Dec 22, 2025

Uh oh!

Uh oh!

Uh oh!

hsliuustc0106 commented Dec 24, 2025

Uh oh!

ZJY0516 commented Dec 24, 2025

Uh oh!

linyueqian commented Dec 24, 2025

Uh oh!

hsliuustc0106 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

linyueqian commented Dec 16, 2025 •

edited

Loading