what are you using to make these?
I am using the following setup, links included if the forum lets me. I am dumbing this down a little for less technical people who might want to try. I have a 12 GB RTX 3080 TI and 32 GB RAM and these run in about 6 minutes per 5 second 720x480 video. There is a workflow and models for Image2Video you can download as well. I don't do too much I2V, but I have had better luck when it comes to hyperbreasts using I2V. It is easier to generate humongous boobs using Pony or Illustrious, then animate them with WAN.
ComfyUI-Easy-Install - This installs a "portable" version of ComfyUI for people with nVidia video cards If you have AMD you will need to find a ComfyUI that works with AMD. You can move the directory around to whatever drive you want. Run the installer and you might want to use SageAttention, it will speed things up during runs.
https://github.com/Tavris1/ComfyUI-Easy-InstallSmooth Mix Wan 2.2, both T2V High and Low v2.0. Download them and put them into the ComfyUI-Easy-Install\ComfyUI\models\unet folder.
https://civitai.com/models/1995784?modelVersionId=2323420https://civitai.com/models/1995784?modelVersionId=2324440If you have 8 GB of VRAM or less, you might need the GGUF versions of the models instead found on this page. I can't recommend any models as it depends on your setup, but you need both the high and the low. The GGUF versions might go into the ComfyUI-Easy-Install\ComfyUI\models\checkpoints folder instead of the unet folder, but I can't say.
https://civitai.com/models/2083588?modelVersionId=2358087Smooth Workflow Txt2Video Workflow v2.0. This is the magic sauce, honestly. Without this workflow you would run out of VRAM. It unloads the models between high/low runs, clears the VRAM, and even upscales the video from a 16 frame small video to 32 frames double the size. Download it and unzip it into your ComfyUI-Easy-Install\ComfyUI\user\default\workflows folder.
https://civitai.com/models/1847730You will also need the following default WAN video components. I cannot tell you how to get them from the WAN repository on GitHub, but if you are techy enough you will figure it out.
ComfyUI-Easy-Install\ComfyUI\models\vae - wan_2.1_vae.safetensors
ComfyUI-Easy-Install\ComfyUI\models\text_encoders - umt5_xxl_fp8_e4m3fn_scaled.safetensors
Once you have them installed/downloaded to the appropriate folders, run run_nvidia_gpu_SageAttention.bat. Once it comes up, click "workflows" on the left bar and select WAN 2.2 Smooth Workflow...
You will get errors that modules are missing. Tell it to install the missing modules. You might have to use Comfy Manager by clicking "Manager" on the top bar and clicking "Install Missing Custom Nodes". I can't handhold through the entire process, but that should download the remaining components you need. Restart Comfy and you will hopefully be good to go. You might have to do this more than once.
Now the workflow will come with the GGUF nodes active. If you have 12GB or more VRAM and are running the standard high/low model you will need to "unbypass" the diffusion nodes in the workflow and bypass the GGUF nodes. You will also have to reconnect the nodes to the LoRA loader for High and Low as well. It should look like the picture I have attached. To bypass/unbypass nodes, just click on the node and click the circle with a line through it (also in the picture attached). They get a purple background when bypassed.
CivitAI has a good selection of LoRAs. Make sure to look for ones for WAN 2.2, you need the high and low noise versions of them. Download both high and low and put them in the ComfyUI-Easy-Install\ComfyUI\models\loras folder. To add LoRAs just click the "+ Add Lora" button in the "HIGH LORA LOADER" node, add the high version, then repeat for the low version in the "LOW LORA LOADER" node. Add your prompt (I recommend leaving the negative prompt as-is, but you can add to it), make sure the size is appropriate (720x480, 480x720, or 640x640), leave the rest at defaults unless you want to push how long the video is, but more RAM/VRAM is needed for longer videos.
BAM! Hopefully in about 6 minutes you have your first high-quality AI video made on your own potato PC!
Quick note, it will create a temp directory in the ComfyUI directory. When you stop the server by doing a CTRL-C in the command window it will clean up the temp directory. All outputs go into the ComfyUI/output directory. It will put them into a "video" folder, then a folder with the current date. Those are left behind and you will need to clean them up periodically if you has low disk space. I use Honeycam to convert them to compressed WEBP animations to embed them into the forum seeing people have issues viewing MP4s.
Punchy