Uploaded January 2026 | Updated November 2026, 3 days ago
In this video, I show you how to generate continuous AI videos longer than 40 seconds with perfect character consistency. Most users struggle with faces melting or motion getting stuck after the first few seconds. I break down my exact "Manual SVI" method that fixes this in just one click.
I walk you through the complete workflow using the Smooth Mix Wan 2.2 model instead of the standard version. You will see why this specific model handles complex motion—like a Tom and Jerry chase scene—much better than the base model. I also explain why you must delete the "Inspector Blend" node if you want your characters to actually move.
If you have a low VRAM GPU, I cover the GGUF settings and the specific "Shift" values you need to change to prevent color distortion in long clips.
Resources & Files You Need:
Model: Smooth Mix Wan 2.2 (High Noise & Low Noise)
civitai.com/models/1995784?modelVersionId=2260110
civitai.com/models/1995784?modelVersionId=2259006
GGUF: huggingface.co/Bedovyy/smoothMixWan22-I2V-GGUF/tree/main
LoRA 1: Rank 64 LightX2V (Dec 17th Version) huggingface.co/lightx2v/Wan2.2-Distill-Loras/tree/main
huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Stable-Video-Infinity/v2.0
LoRA 2: SVI V2 Pro I2V huggingface.co/vita-video-gen/svi-model/tree/main/version-2.0
Extras: Wan 2.2 Text Encoder & VAE
Workflow: aistudynow.com/wan-2-2-svi2-pro-workflow-guide-for-long-ai-videos
#Wan2 #ComfyUI #SVI2Pro #AIVideo
In this video, I show you how to generate continuous AI videos longer than 40 seconds with perfect character consistency. Most users struggle with faces melting or motion getting stuck after the first few seconds. I break down my exact "Manual SVI" method that fixes this in just one click.
I walk you through the complete workflow using the Smooth Mix Wan 2.2 model instead of the standard version. You will see why this specific model handles complex motion—like a Tom and Jerry chase scene—much better than the base model. I also explain why you must delete the "Inspector Blend" node if you want your characters to actually move.
If you have a low VRAM GPU, I cover the GGUF settings and the specific "Shift" values you need to change to prevent color distortion in long clips.
Resources & Files You Need:
Model: Smooth Mix Wan 2.2 (High Noise & Low Noise)
civitai.com/models/1995784?modelVersionId=2260110
civitai.com/models/1995784?modelVersionId=2259006
GGUF: huggingface.co/Bedovyy/smoothMixWan22-I2V-GGUF/tree/main
LoRA 1: Rank 64 LightX2V (Dec 17th Version) huggingface.co/lightx2v/Wan2.2-Distill-Loras/tree/main
huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Stable-Video-Infinity/v2.0
LoRA 2: SVI V2 Pro I2V huggingface.co/vita-video-gen/svi-model/tree/main/version-2.0
Extras: Wan 2.2 Text Encoder & VAE
Workflow: aistudynow.com/wan-2-2-svi2-pro-workflow-guide-for-long-ai-videos
#Wan2 #ComfyUI #SVI2Pro #AIVideo
![Wan 2.1 ComfyUI Text-to-Video & Image-to-Video Workflow [Generate for Unlimited Time]
In this video, you’ll learn:
Generate infinite Time AI videos using the Wan 2.1 ComfyUI workflow?
So I made this workflow that basically lets you generate AI videos nonstop — like, text-to-video or image-to-video, either way. The main trick here is using the For Loop node in ComfyUI. It just repeats the whole generation process however many times you want.
You can set the number of loops manually if you want, or let the math node figure it out based on the video length you’re going for. Either way, it just runs the model a few times and puts it all together at the end.
The loop setup uses the Wan 2.1 model , and if youre going the image-to-video route, theres a toggle for that too. Once the loop runs, the For Loop End node stitches everything together into one smooth output. But here’s the thing — when you use overlapping frames (which helps avoid sudden cuts), some frames get removed. So even if you aim for 15 seconds, you might end up with closer to 12 seconds of usable footage.
I tested this with a prompt like man riding a bike with explosions, and it worked better than I expected. The motion looked natural, and the transitions were smooth thanks to the overlap. Then I changed the prompt in one of the loops to add fire effects around the bike — and it actually picked that up. That said, after the first change, the same visual theme repeated, so it’s best to keep tweaking prompts if you want variety.
If you’re working with image-to-video instead, just disable the text-to-video section and select your input image. Same loop logic applies.
Resources mentioned:
• T-shirt: https://wilddragon.in/product/this-is-my-ai-vibe-oversize-t-shirt-for-men/
Coupon code: aistudynow
• [ComfyUI Wan 2.1 Model]: https://huggingface.co/vrgamedevgirl84/Wan14BT2VFusioniX/tree/main
• [Wan 2.1 VACE Model]: https://huggingface.co/QuantStack/Wan2.1_T2V_14B_FusionX_VACE/tree/main
• Free Workflow: https://aistudynow.com/how-to-create-unlimited-length-videos-with-wan-2-1-comfyui-workflow-2025/
#Wan21 #ComfyUI #AIVideoGeneration #ImageToVideo Wan 2.1 ComfyUI Text-to-Video & Image-to-Video Workflow [Generate for Unlimited Time]](https://i.ytimg.com/vi/Pk-ZzDILOGM/mqdefault.jpg)









