🚀 Ultimate AI Video Production Workflow (ComfyUI)
This is a complete end-to-end AI video pipeline that combines:
🧩 Qwen-Edit 2509 → Image Editing
🎭 Wan Animate 2.2 → Character Animation
📺 SeedVR2 Upscaler → Final 4K Enhancement
Everything is packed into a clean, modular, subgraphed workflow that removes the headache of giant node spaghetti.
This workflow allows you to:
Edit any reference image
Animate the edited character using a real motion video
Upscale and clean the final output
Produce 4K smooth AI videos
Run everything inside ComfyUI with optimized VRAM usage
Perfect for:
AI character reels
Cosplay transformations
Cinematic AI edits
TikTok / IG Reels
YouTube shorts
Film-style previews
🎛️ Features
✨ 1. Qwen-Edit 2509 FP8 (Image Editing Engine)
Outfit changes
Face cleanup / enhancement
Makeup edits
Background removal
Style injection
Very VRAM-friendly
🎬 2. Wan Animate 2.2 FP8 (Character Animation)
Robust identity preservation
Smooth motion from reference video
Pose & face control
Windowed batching for long videos
Subgraphed for easy navigation
📈 3. SeedVR2 Upscaler (Video Restore + 4K)
2× / 4× upscaling
Face enhance option
Removes noise and artifacts
Makes AI videos look real and cinematic
🔧 Bonus Modules
VRAM Cleaner
Resolution Selector
Reference Routing System
Organized Subgraphs
Cleaner wiring = easier editing
🧰 How to Use This Workflow
1️⃣ Load your reference image
Use Qwen-Edit to perform any image cleanup or enhancement before animation.
2️⃣ Load your motion video
Wan Animate will replicate the motion on your edited character.
3️⃣ Adjust resolution & pre-process settings
Resolution presets included (320 → 1280).
4️⃣ Run Wan Animate 2.2
Generates stable character animation frames.
5️⃣ Pass output into SeedVR2
Upscale to 2K / 4K depending on hardware.
6️⃣ Save final video
Both preview & HQ versions included.
🖼️ Qwen Image Edit FP8 (Diffusion Model, Text Encoder, and VAE)
These are hosted on the Comfy-Org Hugging Face page.
Diffusion Model (qwen_image_edit_fp8_e4m3fn.safetensors):
https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/blob/main/split_files/diffusion_models/qwen_image_edit_fp8_e4m3fn.safetensorsText Encoder (qwen_2.5_vl_7b_fp8_scaled.safetensors):
https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/blob/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensorsVAE (qwen_image_vae.safetensors):
https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/blob/main/split_files/vae/qwen_image_vae.safetensors
💃 Wan 2.2 Animate 14B FP8 (Diffusion Model, Text Encoder, and VAE)
The components are spread across related community repositories.
Diffusion Model (Wan2_2-Animate-14B_fp8_e4m3fn_scaled_KJ.safetensors):
https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/blob/main/Wan22Animate/Wan2_2-Animate-14B_fp8_e4m3fn_scaled_KJ.safetensorsText Encoder (umt5_xxl_fp8_e4m3fn_scaled.safetensors):
https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensorsVAE (wan2.2_vae.safetensors):
https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/blob/main/split_files/vae/wan2.2_vae.safetensorsLightx2v : https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v
💾 SeedVR2 Diffusion Model (FP8)
Diffusion Model (seedvr2_ema_3b_fp8_e4m3fn.safetensors):
https://huggingface.co/numz/SeedVR2_comfyUI/blob/main/seedvr2_ema_3b_fp8_e4m3fn.safetensors
🧱 Why This Workflow is Easy to Use
I wrapped the heavy sections inside subgraphs, so you don’t have to scroll through 200 tangled nodes.
Subgraphs included:
Wan Animate Engine
Qwen-Edit Module
SeedVR2 Upscaler
VRAM Cleaner
Resolution Manager
Makes editing, debugging, and expanding the workflow extremely clean and fast.
❤️ Credits
Kijai for FP8 Wan Animate
Comfy-Org for Qwen-Edit
ByteDance for SeedVR2
Community testers & contributors
Description
Added Qwen VL
Added Qwen Image
Added Qwen Image Edit With Masks + Multiple Images
Added SeedVr2 Improvements For Lower Vram
Added FlashVSR1.1
Added Wan 2.2 I2V
Added Wan 2.2 T2V
Added Wan 2.2 FLF2V
Added Wan Animate Improvements For Lower Vram
Easier To Understand Workflow
FAQ
Comments (15)
Does this use Sage Attention? If not, would it benefit from using it?
Speed boost 2x-3x thats the real deal and very low quality drop
Why are you adding a broken workflow with a node 267 error? ‘Mega workflow,’ for fuck’s sake.
267 is a vram cleaner!
@bikiniloversuk456 The issue occurs with every node. I know which number belongs to which one.
I can't figure out if this workflow has Wan I2V. I see Wan T2V, Wan Animate, Wan Painter, and Wan FLF
The painter is only the I2V you check the instructions and also going in the subgraph you can find the node
Kind of a comfy noob - when I open this workflow there are missing nodes obviously. However, Comfy Manager isn't finding them to install. For example, LayerUtility: PurgeVRAM V2, DrawViTPose,SeedVR2VideoUpscaler, etc.
How can I find these? Googling was taking me to documentation
update switch to nightly version of comfyui in the manager and the research im making a tutorial
Most of the URL's in your model list for HuggingFace are returning 404 already.
Do you do any consulting work? I've got some paid work if you are available.
yes please msg me here : https://www.reddit.com/user/Lower-Cap7381/
what are you supposed to do at qwen 4b instruct huggingface page???

