✨ WAN2.2 — Image to video — Simple Workflow
A clean, all-in-one WAN image-to-video workflow built entirely with the UmeAiRT Toolkit for ComfyUI.
Only 12 nodes. No spaghetti wires. Just load your model, write your prompt, and hit generate.
⚠️ IMPORTANT — Nodes 2.0 Required
This workflow is built for the Nodes 2.0 (Vue) interface of ComfyUI. If you don't enable it, the workflow may have display problems.
How to activate Nodes 2.0:
Open ComfyUI
Go to Settings (⚙️ icon, bottom-left)
Find "Use Nodes V2 (Vue)" and toggle it ON
Refresh the page
Load the workflow
If you prefer the classic interface, check out my Legacy version of this workflow instead (link).
🎯 Features
Text-to-Image generation
Automatic download of models in auto version
Built-in SeedVR2 upscaler — high-quality tiled upscaling (toggleable on/off) Slower than a classic upscaler, but significantly better quality
Full metadata embedding — your images are saved with all generation parameters, ready for online publishing and remixing
3 LoRA slots — with individual on/off toggles and strength control and you can connect as many other lora modules to each other for as many LoRA as you want.
📦 Custom Node Required
Only one custom node to install:
Install via ComfyUI Manager (search "UmeAiRT") or use the UmeAiRT Auto-Installer.
The Toolkit packages everything internally — upscaler, face detailer, metadata saver. No other custom nodes needed.
📂 Files you need (in manual version)
For base version
I2V Model : wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors and wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors
In models/diffusion_models
CLIP: umt5_xxl_fp8_e4m3fn_scaled.safetensors
in models/clip
For GGUF version
I2V Quant Model : wan2.2_i2v_high_noise_14B_QX.gguf and wan2.2_i2v_low_noise_14B_QX.gguf
In models/unet
Quant CLIP: umt5-xxl-encoder-QX.gguf
in models/clip
For speed version
lightning LoRA : Wan2.2-Lightning_I2V-A14B-4steps-lora_HIGH_fp16.safetensors and Wan2.2-Lightning_I2V-A14B-4steps-lora_LOW_fp16.safetensors
For AIO version
I2V FAST AIO Model : wan2.2-i2v-rapid-aio.safetensors
In models/checkpoint
VAE: wan_2.1_vae.safetensors
in models/vae
ANY upscale model:
Realistic : RealESRGAN_x4plus.pth
Anime : RealESRGAN_x4plus_anime_6B.pth
in models/upscale_models
Description
Beta version
FAQ
Comments (46)
Have updated Comfy and all the modules, but sadly I keep getting a "Text Find and Replace" missing node. Even its all updated
EDIT: Just disabled the automatic prompt and removed those nodes. But not triton got borked so I'm just re-installing Comfy using your script
EDIT2: Comfy installed with your script runs this workflow with no issue! Yay!
Sweet I gotta do a fresh ComfyUI setup, so many tweaks over the months. Thanks for posting results.
EDIT 3: Yay learn to read verbose
TeaCache? Really? no-one should be using this atrocious quality destroying concept on video generation- especially in the age of perfect speed-up LoRAs. The problem with AI is anyone can propose anything, and some truly awful ideas gain traction and become things of legend that too many people want without knowing why the idea was so awful in the first place.
Recommended samplers/schedulers are another. So much output is substandard, because people parrot the worst samplers/schedulers as they simply copy what they've seen in other workflows.
I would say what I see as 'common knowledge' in local AI is almost always wrong, and often wrong to a scary degree. Wan2.2 is a fantastic model, and it is sad that people with bad workflows honestly think there is little improvement over 2.1.
it's too easy to talk. Show us something better.
you can just disable it. the man is providing great workflows. with customization not forcing you to use it.
Bro sounds like a groomer having a meltdown. LEA must love you.
waow
Second workflow in the last 30m that I see this user bashing the poster but all empty worlds so far.. especially the "perfect speed loras" 😂
Can you check please and build workflow with LORA for this All in one WAN 2.2 model merges: 4-steps, 1 CFG, 1 model speeeeed (both T2V and I2V) https://www.reddit.com/r/StableDiffusion/comments/1mddzji/all_in_one_wan_22_model_merges_4steps_1_cfg_1/
Thank you!
I agree 👍🏼 I tested the AIO model from reddit and it's great. I adapted an old version of this workflow to use it.. one dedicated to the AIO model would be awesome!
Ive been having good results with just replacing the model, clip and vae loaders with the checkpoint loader in the example provided, just three connections that need to be redone.
I just added a version to use this model
What's the difference between Lora - High Noise and Low noise ?
Also, it's GGUF version supposed to be faster than the normal one ? Or generation time it's like same ?
I'm on a 5090 , everytime i used the old 2.1 normal wokflow, i almost got no movement at all compared with the gguf version :S
Thank you so much for your workflows :D
The current operation of WAN2.2 uses two model files, so loras must be applied to both.
The GGUF version is of lower quality and generally slower than the basic version, it is only useful for reducing VRAM consumption.
same loras to both ? Gonna give a try maybe that's the problem i was encounternig :D
UPDT 20:00
As an early version it works great even tho slower cause the high and low sampler but movements it's impressive :D
Cannot wait for future version with preview too maybe :)
Thank you for your work as always
UmeAiRT Regarding the quality a Q8 gguf is superior to a fp8scale. Fe https://www.reddit.com/r/StableDiffusion/comments/1eso216/comparison_all_quants_we_have_so_far/ also from my testings and conversions
is it compatible with yours "ComfyUI auto installer FLUX | GGUF | CONTROLNET | UPSCALE" ? It gives me error "TorchCompileModelWanVideo
Failed to compile model"
Just unchek "torchCompile" optimisation
Can you help me with this node?
It's from your SDXL Controlnet workflow
https://ibb.co/W4xsjy9d
I can't seem to find what pack it's from or what name it has, comfyui does not detect it.
I want to be able to save my loras and their weight properly on my workflow but I've been stuck for months not being able to add lora info to my image saver...
This is my workflow
https://www.mediafire.com/file/kcrgh90wt5odvkj/Entice+SPARK.json/file
I started it off this one you uploaded a while back.
https://civitai.com/models/1379025/img-to-img-simple-workflow-illustrious-sdxl-or-upscale-or-lora-or-civitai-metadata
If someone on the wf "classic" gets pixel noise when generating with lora LightX2v enabled, then change the value in the sampler
https://civitai.com/models/1824594?modelVersionId=2064786
Works better with your workflow, IMHO
can u do 2.2 (Fast) with First frame - Last Frame? Pls...
I will try to publish it soon
I added the UnetMultiGPu and it works well. Thanks for the great workflow as always.
You have a blockswaping module already in the workflow
I love your work man. I just wanted to ask... is it just me, or the model wants to end the clip with the same image it started with?
i've been having the same issue with a different workflow where i swear it's trying to loop
if you are exceding the time like going to above 8 seconds the light lora does that it wants to return to original one. check the frames for the output video if you do 16 you can do more seconds play with speed and make sure to check the node multiplier for the duration that alos match de 16 output .
fadog74502638 Thanks for the info. I usually don't push over 5s. I didn't quite catch the second part you wrote though. I'm not a native speaker, sorry :/
Korialstrasz at the bottom of the workflow you need to expand the node called "calculFrames", the top entry for "b" has to match the framrate of the output (the top entry on the node where your generated clip is displayed at). So type 16 there and 16 in the expanded calculFrame-node.
Looking forward to trying this;
Have you thought of integrating MagCache? its superior to TeaCache. I was using it with your workflow in place of teachache prior to discovering wan2.1 4 step fast loras, but skipping steps with only 4 steps is not wise. https://zehong-ma.github.io/MagCache/
You can find it in ComfyUI Custom Nodes Manager;
https://github.com/Zehong-Ma/ComfyUI-MagCache
I have very bad quality with magcache it's why i never use it, maybe it's better now i give it another chance
If someone on the wf "classic" gets pixel noise.The issue lies with the Sampler's steps and start_at_step parameter settings, which caused the low sampler to fail.
anyone getting videos with jaggies? the image is kind of pixely & crunchy.
Like someone scaled down an image with "nearest neighbor".
I get this when the output resolution is to low. If I use 480p and higher, it looks very clean.
Awesome work like your older Wan2.1 i2v wf!
Generated a 5sec clip (1280x720) in 426 secs (7min)
Activated every setting except teacache and upscale/interpolation.
Hardware: RTX4080 (16GB VRAM) / 32GB DDR4
Can any 2.1 lora work with 2.2 model
I try some it can't not load
lora key not loaded: diffusion_model.blocks.9.cross_attn.v_img.lora_B.weight
Tested and it work.
I found this, seems like the Wan 2.1 I2V loras have extra/more blocks that 2.2 does not use but sounds like it can use the remaining blocks? Also sounds like Wan 2.1 T2V does not have these extra blocks so it should not error? I dunno, I only started using Wan2.2 today and I see that error too. https://www.reddit.com/r/comfyui/comments/1mgo0z3/wan_22_doesnt_load_certain_loras_i_got_it_working/