Wan21_VACEV2V_Wrapper_LongLooping
Generates long looping videos with Wan2.1 using the VACE wrapper, optimized for low VRAM stability.
Wan21_VACEV2V_LongLooping
A simpler version for Wan2.1 V2V (video-to-video), focused on long seamless loops.
Upscale_video
Enhances video resolution and quality by applying AI upscaling models.
T2V_Wan21
Converts text prompts into videos using Wan2.1 (text-to-video generation).
Nunchaku_Upscale_Image
Upscales still images with Nunchaku, preserving details while increasing resolution.
Nunchaku_FluxKontext_Image
Generates or edits images with Flux.1 using Kontext prompts for precise control.
Nunchaku_FluxCN_Image
Uses ControlNet with Flux.1 inside Nunchaku to apply structure (pose, depth, etc.) to images.
Nunchaku_Flux_Image
A standard image generation workflow with Flux.1 inside Nunchaku, optimized for low VRAM.
I2V_InfiniteTalk_Wan21
Creates talking or animated videos from input images using Wan2.1 with voice embeddings (InfiniteTalk).
FunControl_V2V_Wan21
A flexible video-to-video workflow for Wan2.1, with added control features (depth, pose, etc.).
FlF2V_Wan21
Generates a smooth video transition from a first frame to a last frame using Wan2.1.
Description
Used of Nunchaku and long video with VACE
FAQ
Comments (8)
Hey @Akumetsu971 . I have one question - How to improve quality -> I mean If I wre to use on a high vram setup - WHat do I need to change (apart from models and offloading)
Or Do I have to use some other nodes?
Just wanted to make this workflow for let's say 48VRAM
Easier to use another workflow- there are so many on this site. Anyway, in general, render to as high a resolution you can BEFORE any upscale nodes. Use initial iterations at a high CFG (>1) to lay down decent motion. Investigate the use of bong_tangent and bong maths (google is your friend).
But as I said, it is easier to just search for a more appropriate workflow for your 48GB VRAM here or on Reddit.
@blobby99 I tried a lot and must say this one works like charm. If you happen to know where I don't have to extract frames in chunk then combine etc - Please share that template which you personally know worked well.
To improve, you should remove the gguf nodes in the model node group and use the regular one. Also, you may want to use the wrapper wf where you just replace the gguf model by fp16 or fp8 models. And change the block swap parameters at 3-4. Otherwise, I dont see what higher setup do different than me ?
@Akumetsu971 he means what i mean: wan 2.2 : )
I tried the "Wan21_VACEV2V_Long Looping" workflow and was pleasantly surprised. You've solved a lot of WAN problems. The results are without frame loss, and there are no blurry frames either. Great work. Can you also transfer this to WAN 2.2 VACe? Here are the models: https://huggingface.co/lym00/Wan2.2_T2V_A14B_VACE-test/tree/main
Really easy to modify for wan2.2. You only need to adjust the two sampler groups and the model group. I thought to use wan2.2 but it is more complicated on low vram
@Akumetsu971 no, its not easy, i have try it the last 10 hours without luck.