Timeline for WAN2.1 Jenga + ComfyUI Support?

[EladWarshawsky]

Hi, I appreciate you open-sourcing the project, it’s been great to dig into. I noticed WAN2.1 and ComfyUI support are both on the roadmap. Do you have any rough timeline or priority order for these features? I would love to see the speedups. Thanks again for the awesome work!

[JulianJuaner]

Thank you for your attention to Jenga! We will complete the open-source adaptation for Wan2.1-1.3B within a short period (this week). Regarding ComfyUI, as I currently lack relevant experience, it may take some time to learn, and I do not have a fixed timeline yet, but we will do our best to expedite it. We also look forward to the community's help achieving a ComfyUI version of Jenga.

[kabachuha]

I think the community (kijai and native comfy) will pick up the implementation quickly from their side, when there will be one (esp. since their implementation differ greatly from vanilla Wan)

[MeiYi-dev]

Imagine this with the AccVideo model that was just released. 30s - 1m for a 480p 5sec video sounds so good. I think ComfyUI implementation would be fantastic.

[kabachuha]

Am I right that the only changes needed are:

build_multi_curve, and hilbert_order (+ subfunctions like gilbert_mapping)?

[JulianJuaner]

@MeiYi-dev That sounds good. We have tested it on 720P. AccVideo shares the same architecture as HunyuanVideo. We have uploaded a sample script to infer on AccVideo.

We also updated a dev-version of Wan2.1-1.3B (Jenga-base / turbo).

[JulianJuaner]

The future plan includes the Wan 14B support, SageAttention, and possible ComfyUI

[MeiYi-dev]

The future plan includes the Wan 14B support, SageAttention, and possible ComfyUI

Any updates regarding ComfyUI support?