🎯 Objective
Evaluate the video quality, render time, and practical workflow implications of the new WAN21-VACE model when used with ComfyUI, focusing on how resolution affects quality and performance.
🧩 Components Used
WAN21-VACE model (native support via ComfyUI update)
🔗 Comfy Blog AnnouncementComfyUI (latest version with VACE native support)
Three resolution settings:
512x768 (high quality)
400x600 (medium test)
256x368 (low resolution + upscale test)
🔬 Explanation
I ran a comparative test by generating three short video clips using the WAN21-VACE model at various resolutions:
512x768 – The highest resolution I tested. The quality is excellent, with crisp details and fluid motion. However, this came at a cost: only 2 seconds of video took a considerable amount of time to render, making it impractical for longer clips unless used selectively for hero shots or key sequences.
400x600 – This resolution strikes a balance between quality and performance. While there's a slight drop in sharpness, the video still maintains respectable detail and coherence. Render time was significantly reduced compared to the 512x768 test.
256x368 + Upscaling – Surprisingly, even at this low resolution, the video retained enough visual information to remain watchable after upscaling. The quality drop was noticeable but not as dramatic as expected, which makes this a viable option for quick drafts or previews.
📦 Attached Video Examples (video samples.zip)
🔹 Video 1 – 512x768 (2s, high quality)
🔹 Video 2 – 400x600 (faster render, medium quality)
🔹 Video 3 – 256x368 (fast render + upscale test)
🧪 ComfyUI Workflow Example
To help you reproduce or adapt my test, I’m including:
Workflow screenshot to visually demonstrate the node setup.
ComfyUI .json file attached, so you can load it directly and experiment with your own settings.
This workflow includes:
Model loading node for WAN21-VACE
Custom resolution input (tested at 512x768, 400x600, and 256x368)
The attached example has also reactor for the face and upscaling node
🧠 Conclusion
The WAN21-VACE model delivers exceptional quality at high resolution, but you'll need patience or serious GPU power. That said, reducing the resolution—even drastically—doesn’t entirely ruin the visual quality, especially when paired with intelligent upscaling. This makes the model flexible for both quick concept testing and final production shots, depending on your time and resource constraints.
It's also worth noting that Hunyuan remains a faster option for generating longer videos with comparable visual quality. While WAN21-VACE excels in detail and motion fidelity, Hunyuan offers a more time-efficient workflow for longer content.
ComfyUI continues to be a robust playground for video AI experimentation, and WAN21-VACE is a powerful addition for creators.