传统工作流直接使用Wan2.1_I2V_14b模型生成720p视频，存在耗时长的问题。
本工作流采用分阶段采样策略优化视频生成效率，速度提升到2倍以上。
第一步：使用Wan2.1_I2V_14b模型生成228p基础视频。
第二步：通过Wan2.1_fun_1.3b模型进行放大，最终输出720p高清视频。
附件中PNG图片隐藏了工作流信息。只要将它拖入Comfyui界面中即可。
Original Workflow:
The conventional workflow directly utilizing the Wan2.1_I2V_14b model for 720p video synthesis suffers from prolonged processing times.
Optimized Workflow:
This pipeline implements a two-stage sampling strategy to enhance video generation efficiency, reducing the total processing time by over 50%:
Base Video Synthesis: The Wan2.1_I2V_14b model first generates a 228p foundational video sequence.
Resolution Enhancement: The Wan2.1_fun_1.3b model then performs temporally consistent super-resolution upscaling to deliver 720p high-definition output.
The workflow information is hidden in the PNG image attached. You just need to drag it into the ComfyUI interface.
Wan2.1-I2V-14B-480P-gguf:https://huggingface.co/city96/Wan2.1-I2V-14B-720P-gguf/tree/main
Wan2.1-Fun-1.3B-Control: https://huggingface.co/alibaba-pai
text_encoders+clip_vision+vae:https://blog.comfy.org/p/wan21-video-model-native-support
Description
FAQ
Comments (15)
太好了，正在找，感谢大佬
Is the lora used in the workflow important?
If it is, can you point to where we can get it?
This lora is not necessary.
For a 5 sec video, how long will it take the time to generate?
also can i stack the lora?
Tinkered some time with it.
Very nice idea, noticed some things
1) It can help with some loras that can't do a crisp image.
2) It really is noticeably faster than one-stage workflow.
3) But couldn't make a consistent generation in i2v with 6 steps, changed to 12 with 6/6.
4) In the v2v you have batch size set to 2. It's should be 1.
Thanks for testing the workflow! Can't wait to hear more from you.
got stuck on vae decode. i have 12gb vram. 64gb ram. what settings or node should i change to make it work?
It can run even with 8GB VRAM.Perhaps there is a version mismatch with your VAE.You can download the required VAE files from the webpage below.
@Xi_mu yeah i already used that vae, and still stucks. any idea?
btw where you download the wan2.1-fun-1.3b-control model?
@DarkAmbassador Wan2.1-Fun-1.3B-Control https://huggingface.co/alibaba-pai
@Xi_mu now it runs but the result is blurry and bad, not sure why :/
@DarkAmbassador I'll need to adjust some parameters. I'll release Version 2 of the workflow when I have time.