Update 8/12/2025: SwarmUI fixed an issue
Previously, the LORAs applied after <videoswap> weren't actually being applied. Updating workflow below to reflect the adjusted approach.
Capturing a quick guide for posterity on how my newbie self managed to get Wan 2.2 working "decent-ish" for I2V in SwarmUI (here's the T2V version). Thanks to the guides and builders here and all around the internet for all the bits and pieces I managed to cobble together.
Getting 81 frames at 16fps (best looking motion at baseline for now) in < 3 minutes total (high and low pass). I'm on an RTX 3090 (24GB VRAM) with 128GB system RAM (don't need that much for this).
I'm POSITIVE I could get better results than this if I knew more than I do, but it's working and I'm content with that at this early stage in the Wan 2.2 lifecycle. Hell, I've only been playing with this stuff for 3 weeks.
SwarmUI preset attached.
Models and LORAs
This works with pretty much any of the wan 2.2 high and low combos. I've tried it with the fp8 safetensors and the Q4_K_M and Q5_K_M GGUFs found all over the place.
In Swarm, I do change the model resolution for both high and low in the metadata from 960x960 to 640x640 for speed.
The key LORA for my setup is the wan 2.1 lightx2v version (I'm using this one: Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64). Key here is to set it to 1.5 strength.
The other causvid/self-forcing/blah blah LORAs may or may not work fine here, too. Haven't tried them yet (I was so excited to get this functioning I quit messing around.)
The 2.2 high/low lightning LORAs do work, but they still seem to affect the motion more than the 2.1 lightx2v LORA. Experiment with both.
For easy preservation, I saved my preset with the lora tag built into the prompt like this:
{value}
<video>
<lora:video/control/Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64:2.5>
<videoswap>
<lora:video/control/Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64:1>(Hat tip to blobby99 for the tip on lora weights; I still experiment with 2.5:1 vs 1.5:1.5)
That's how it'll look in your preset. If copy/pasting, just change {value} to whatever prompt you want.
SwarmUI Setup
This was the tricky part to get together. I'll go through the key settings here, and they should all be captured in the attached preset.
Prompt: optionally include the lightx2v LORA above
Sampler: Euler
Scheduler: Simple
Init image: pick your pic
Init image creativity: 0 as always
Video model: pick your version of the HIGH wan 2.2 model
Video swap model: pick your version of the LOW wan 2.2 model
Video swap percent: 0.5
Video frames: 81 (for 5 seconds; or your multiple of 16 + 1 frame)
Video steps: 4
Video CFG: 1
Video Resolution: Image aspect, model resolution (see model resolution note above)
Video FPS: 16
Sigma Shift: 5
That's basically it. I've played with other samplers/schedulers and honestly can't tell a difference. But I'm not aiming for cinema quality shots.
Bonus Tip
If you want to play with the high and low lightning (or any other high/low pair) loras, here's what you add to the END of your prompt in SwarmUI to get them to apply at the right times:
<video>
<lora:video/wan22/Wan2.2-Lightning_I2V-A14B-4steps-lora_HIGH_fp16:2.5>
<videoswap>
<lora:video/wan22/Wan2.2-Lightning_I2V-A14B-4steps-lora_LOW_fp16:1>I'd just type them in. As you begin typing <lora: you'll get a popup with a list of your LORAs. Makes it super easy to apply them at the correct stages (and also easy to save settings you like in presets).
Wrapping Up
I THINK I've captured the most important bits and pieces here. I'll try to help if anyone has a question, but I'm still pretty lost myself on most of this.
G'luck!