✨One-click Pod available on:✨

🟣 Deploy on RunPod with CUDA 13.0

🟣 Deploy on RunPod with CUDA 12.8

🟡 Deploy on VastAI

🐳 RunPod users: Just click the template link, choose a GPU, and everything installs automatically — ComfyUI, all nodes, all workflows, and WAN 2.2 models (~30GB) download in the background on first boot. No manual setup needed. ComfyUI starts immediately while models download.

☕️ buymeacoffee

IMPORTANT:

If you install RES4LYF node it will broke the MoEKSampler, to use it you have to use the KSampler included in that node.

ComfyUI-QwenVL-Mod — Enhanced Vision-Language with WAN 2.2 Version 2.9.0 (2026/09/07) — 🎬 WAN 2.2 NSFW Video + WAN Remix T2V/I2V Models + Story/Timeline Workflows (up to 20s) + SVI Camera + FL2V First-Last-Frame + 8 Workflows + Wildcards Included

⬆️ 2026/09/07 UPDATE ⬆️

📦 What's Included — 8 Workflows

All workflows are pre-wired with Qwen3-VL auto-prompting, WAN Remix diffusion models, and TensorRT upscale + RIFE interpolation where applicable.

• WAN2.2-T2V-Qwen3.5.json — T2V · Text-to-video, 5 seconds

• WAN2.2-I2V-Qwen3.5.json — I2V · Image-to-video, 5 seconds

• WAN2.2-FL2V-Qwen3.5.json — FL2V · First-Last-Frame to video, TensorRT upscale + RIFE

• WAN2.2-I2V-20s-Qwen3.5.json — I2V 20s · Single-scene image-to-video, 20 seconds

• WAN2.2-I2V-20s-Story-Qwen3.5.json — Story I2V · Multi-prompt timeline, 20 seconds (4 × 5s)

• WAN2.2-I2V-SVI-20s-Qwen3.5.json — SVI 20s · Subject Video Identity, 20 seconds

• WAN2.2-I2V-SVI-20s-Story-Qwen3.5.json — Story SVI · Timeline with SVI identity lock, 20 seconds

• WAN2.2-T2V-I2V-Story-Qwen3.5.json — Story T2V+I2V · Timeline mixing T2V and I2V, 20 seconds

🔄 WAN Remix T2V/I2V Models — 4 Variants

All WAN 2.2 workflows now use the WAN Remix T2V/I2V diffusion models. Download from the original Civitai pages:

• wan22RemixT2VI2V_t2vHighV20.safetensors — T2V · High motion · ~14.3 GB — Civitai — WAN REMIX T2V v2.0 High

• wan22RemixT2VI2V_t2vLowV20.safetensors — T2V · Low motion (stable) · ~14.3 GB — Civitai — WAN REMIX T2V v2.0 Low

• wan22RemixT2VI2V_i2vHighV30.safetensors — I2V / FL2V / SVI / Story · High motion · ~14.3 GB — Civitai — WAN REMIX v2.1 FP8 High

• wan22RemixT2VI2V_i2vLowV30.safetensors — I2V / FL2V / SVI / Story · Low motion (stable) · ~14.3 GB — Civitai — WAN REMIX v2.1 FP8 Low

• Credits: FX_FeiHou (FP8 Remix)

• High vs Low: High = more dynamic camera and subject motion; Low = more stable, controlled motion (better for subtle animations)

🧹 Removed: WAN Enhanced NSFW SVI Camera

• Removed wan22EnhancedNSFWSVICamera_nsfwV2FP8H/L models — superseded by WAN Remix

• Docker and provisioning cleaned up

🎲 PMP Wildcards — Downloaded at Boot

• Wildcards (__pmp/prmpt/*) are now downloaded from ComfyUI-Garage at boot time

• No Docker rebuild needed to update wildcards — just push to Garage and restart the pod

• comfy-tagcomplete ships with wildcard fallback for local installs

⬆️ 2026/08/04 UPDATE ⬆️

✨ ComfyUI QwenVL-Mod Node Update ✨

v2.4 — Local Model Discovery + Qwen3.5 + SageAttention

We haven't forgotten about this node! Here's what's new since v2.2:

• 🎬 LTX 2.3 Presets (v2.3): New specialized presets for LTX 2.3 I2V and T2V with official prompting guides. Multilingual support for all presets, simplified single-paragraph format (max 200 words), full NSFW support.

• 🔍 Local Model Discovery (v2.4): Drop your GGUF/HF files into models/LLM/ and they show up in the dropdown automatically — no more JSON editing. Auto-pairs mmproj files for vision GGUF models.

• 🧠 Qwen3.5 support (v2.4): Architecture detection from file metadata (GGUF header / HF config.json), automatic thinking-mode disabling, forced top_k=20.

• ⚡ SageAttention restored (v2.4): Architecture-aware kernels (Blackwell FP8, Hopper FP8, Ada FP8, Ampere FP16) with graceful SDPA fallback.

🔧 Also updated our companion nodes:

• ComfyUI-Upscaler-TensorRT-Auto — TensorRT upscaling with auto-detection, CUDA 12/13 wheels pre-baked

• ComfyUI-RIFE-TensorRT-Auto — TensorRT frame interpolation, CUDA 12/13 wheels pre-baked

• comfy-tagcomplete — Tag completion with wildcard support for WAN 2.2 workflows

• ComfyUI-HuggingFace — Model download integration for local discovery

All WAN 2.2 and LTX 2.3 workflows (T2V, I2V, SVI, MMAudio, GGUF variants) are tested and working with the updated nodes. Grab the latest version and let us know how it goes!

⚠️ Requirements — Read First!

GPU & VRAM

• 🟢 Recommended — RTX 5090 (32 GB) / RTX PRO 6000 (48 GB) / RTX 4090 (24 GB) → FP8 Remix models

• 🟡 Mid-range — RTX 3090 (24 GB) / RTX 4080 (16 GB) → FP8 with offload

• 🟠 Lower VRAM — 12–16 GB → FP8 with aggressive offload

Model Quantization Options

• FP8 (recommended) — ~14.3 GB per diffusion model + ~4.8 GB text encoder = ~19 GB active set → huchukato/garage

• FP16 (full) — ~42 GB per diffusion model + ~12 GB text encoder = ~54 GB total → Comfy-Org/Wan_2.2

Software

• ComfyUI: v0.31.0+

• Python: 3.10+

• CUDA: 12.8+ (13.0 recommended)

• Storage: allow at least 80 GB for the complete provisioned package

Text Encoder

• FP8 (recommended, NSFW): nsfw_wan_umt5-xxl_fp8_scaled.safetensors (~4.8 GB) — NSFW-API/NSFW-Wan-UMT5-XXL

• FP8 (standard): umt5_xxl_fp8_e4m3fn_scaled.safetensors — Comfy-Org

🌟 What is ComfyUI-QwenVL-Mod?

A powerful enhanced vision-language node for ComfyUI that combines Qwen3-VL models with WAN 2.2 video generation workflows. Features multilingual support, visual style detection, NSFW capabilities, Story/Timeline multi-prompt generation, and MMAudio integration.

Think: "Your all-in-one solution for intelligent prompt enhancement and video generation with WAN 2.2!"

🎬 Key Features

🚀 WAN 2.2 Video Generation

• T2V (Text-to-Video): Generate video from text prompts

• I2V (Image-to-Video): Animate a first-frame image

• FL2V (First-Last-Frame): Generate the transition between two keyframes — Qwen3-VL sees both frames

• SVI (Subject Video Identity): Lock character identity across generations using reference images

• Story (Timeline): Multi-prompt timeline generation — up to 4 prompts for 20-second videos with automatic scene transitions

🧠 Qwen3-VL Auto-Prompting

• Multilingual: Write your prompt in any language — Qwen3-VL translates and converts it

• Auto-format: Generates optimized WAN 2.2 prompt format

• Multi-reference: Qwen3-VL sees all connected images via image + image2 inputs

• Visual style detection: 12+ artistic styles (photorealistic, cinematic, anime, 3D CG, claymation, vintage film, watercolor, fantasy, etc.)

• Smart caching: Performance optimization with Fixed Seed Mode

• GGUF backend: Efficient local model inference with quantization support

• Qwen3.5 support: Thinking mode disabled via /no_think for fast prompt generation

• Camera tag dropdown: 19 camera movements selectable directly in the node UI

🎵 MMAudio Integration

MMAudio can be added to any workflow by connecting the MMAudio nodes to the generated video output. The node analyzes the video and produces synchronized audio (music, speech, sound effects).

🎨 NSFW Support

• Comprehensive content generation without restrictions

• Dedicated NSFW presets for each workflow type

• Natural progression, style adaptation, consistent characters

🎯 QwenVL-Mod NSFW Presets

The workflows include built-in NSFW presets for the Qwen3-VL prompt enhancer:

🍿 T2V Presets

• 🍿 Wan 2.2 NSFW T2V — Standard T2V prompt

• 🍿 Wan 2.2 NSFW T2V Timeline (5s) — Timeline format for Story workflows

🎥 I2V Presets

• 🎥 Wan 2.2 NSFW I2V Scene (5s) — Single scene, 5 seconds

• 📖 Wan 2.2 NSFW I2V Scene (20s) — Single scene, 20 seconds

• 🎬 Wan 2.2 NSFW I2V Timeline (20s) — Multi-prompt timeline, 20 seconds

🔄 FL2V Presets

• 🔄 Wan 2.2 NSFW FL2V Scene (5s) — Transition between first and last frame

🖼️ Utility Presets

• 🖼️ Detailed Description — SFW detailed scene description (for non-NSFW use)

SFW presets are also available. Edit the preset dropdown in the QwenVL node to switch.

🖼️ Multi-Reference Input (image2)

The QwenVL-Mod node has two image inputs:

• T2V: no images needed

• I2V: image = first frame

• FL2V: image = first frame, image2 = last frame

• SVI: image = primary reference, image2 = additional references (batch)

• Story: image = first frame for I2V segments, image2 = optional second reference

Qwen3-VL sees all connected images as individual images, enabling proper multi-reference analysis.

🎮 Usage Examples

Basic Text-to-Video (T2V)

1. Load WAN2.2-T2V-Qwen3.5.json

2. Write your prompt in any language

3. Select preset 🍿 Wan 2.2 NSFW T2V

4. Generate video

Image-to-Video (I2V)

1. Load WAN2.2-I2V-Qwen3.5.json

2. Upload your first-frame image to image

3. Select preset 🎥 Wan 2.2 NSFW I2V Scene (5s)

4. Write what happens next (in any language)

5. Generate animated video

First-Last-Frame (FL2V)

1. Load WAN2.2-FL2V-Qwen3.5.json

2. Upload first-frame to image, last-frame to image2

3. Select preset 🔄 Wan 2.2 NSFW FL2V Scene (5s)

4. Describe the transition between the two frames

5. Generate the interpolated video with TensorRT upscale + RIFE

Story / Timeline (I2V Story)

1. Load WAN2.2-I2V-20s-Story-Qwen3.5.json

2. Upload first-frame to image

3. Select preset 🎬 Wan 2.2 NSFW I2V Timeline (20s)

4. Write prompts for each timeline segment (up to 4 prompts, 5s each)

5. Generate a 20-second video with automatic scene transitions

6. Recommended: max_tokens = 2048, context_length = 16384+ for 20s timelines

20-Second Single Scene (I2V 20s)

1. Load WAN2.2-I2V-20s-Qwen3.5.json

2. Upload first-frame to image

3. Select preset 📖 Wan 2.2 NSFW I2V Scene (20s)

4. Write what happens next (in any language)

5. Generate a single-scene 20-second video

SVI — Subject Video Identity (20s)

1. Load WAN2.2-I2V-SVI-20s-Qwen3.5.json

2. Upload primary reference to image, additional references to image2

3. Select preset 🎥 Wan 2.2 NSFW I2V Scene (20s)

4. Generate a 20-second video with locked character identity

Story SVI — Timeline with Identity Lock (20s)

1. Load WAN2.2-I2V-SVI-20s-Story-Qwen3.5.json

2. Upload primary reference to image, additional references to image2

3. Select preset � Wan 2.2 NSFW I2V Timeline (20s)

4. Write prompts for each timeline segment

5. Generate a 20-second Story video with consistent character identity

🔧 Technical Specifications

⚡ Performance

• Output: 720p/1080p, 16 fps (native), up to 20 seconds (Story)

• Upscale: TensorRT RealESRGAN (FL2V workflow)

• Frame interpolation: RIFE v4.25 → 48 fps (FL2V workflow)

• Sage Attention: FP16 accumulation, async offload

• Smart caching: Reuse prompts with same inputs, Fixed Seed Mode for text-only caching

🎨 Model Support

• Qwen3-VL 4B: 7 GGUF variants (2.38 GB – 4.28 GB)

• Qwen3-VL 8B: 7 GGUF variants (4.8 GB – 8.71 GB)

• Qwen3.5: 4B / 9B / 27B (uncensored, heretic, unsloth) — thinking mode disabled

• HF Models: Josiefed, official, Heretic-Stable variants

• Quantization: Q4_K_S, Q5_K_S, FP16, INT8, FP8

🌐 Multilingual Capabilities

• Input languages: Any language supported

• Auto-translation: Automatic translation to optimized English

• Style detection: Works with multilingual prompts

• Cultural adaptation: Context-aware prompt enhancement

📦 Installation

Quick Install

1. Download: ComfyUI-QwenVL-Mod (latest version)

2. Extract to ComfyUI/custom_nodes/ComfyUI-QwenVL-Mod

3. Install requirements: pip install -r requirements.txt

4. Restart ComfyUI

5. Load included workflows from wan22/ folder

Custom Nodes Required

• ComfyUI-QwenVL-Mod — All workflows (Qwen3-VL prompt enhancer) — huchukato/ComfyUI-QwenVL-Mod

• ComfyUI-RIFE-TensorRT-Auto — FL2V (frame interpolation) — huchukato/ComfyUI-RIFE-TensorRT-Auto

• ComfyUI-Upscaler-TensorRT-Auto — FL2V (upscaling) — huchukato/ComfyUI-Upscaler-TensorRT-Auto

• ComfyUI-VideoHelperSuite — All workflows (VHS_VideoCombine) — Kosinkadink/ComfyUI-VideoHelperSuite

• ComfyUI-Easy-Use — FL2V (easy showAnything) — yolain/ComfyUI-Easy-Use

• ComfyUI-PerfectVideoResolution — All workflows (resolution selector) — huchukato/ComfyUI-PerfectVideoResolution

• ComfyUI-WanMoeKSampler — Story workflows (WanMoe advanced sampling) — stduhpf/ComfyUI-WanMoeKSampler

• ComfyUI-PainterI2V — Story workflows (Painter I2V) — princepainter/ComfyUI-PainterI2V

• ComfyUI-PainterLongVideo — Story workflows (Painter long video) — princepainter/ComfyUI-PainterLongVideo

• ComfyUI-mxToolkit — Story workflows (mxSlider) — Smirnov75/ComfyUI-mxToolkit

• ComfyUI-TagComplete — Wildcards (WildcardProcessor + __pmp/prmpt/*) — huchukato/comfy-tagcomplete

• rgthree-comfy — Power Lora Loader, Fast Groups Bypasser — rgthree/rgthree-comfy

• Euler-Smea-Dyn-Sampler — Alternative samplers — Koishi-Star/Euler-Smea-Dyn-Sampler

Models Required

FP8 Workflows (T2V):

• models/diffusion_models/ → wan22RemixT2VI2V_t2vHighV20.safetensors (~14.3 GB) or wan22RemixT2VI2V_t2vLowV20.safetensors — huchukato/garage

• models/text_encoders/ → nsfw_wan_umt5-xxl_fp8_scaled.safetensors (~4.8 GB) — NSFW-API/NSFW-Wan-UMT5-XXL

• models/vae/ → wan_2.1_vae.safetensors (~253 MB) — Comfy-Org

FP8 Workflows (I2V / FL2V / SVI / Story):

• models/diffusion_models/ → wan22RemixT2VI2V_i2vHighV30.safetensors (~14.3 GB) or wan22RemixT2VI2V_i2vLowV30.safetensors — huchukato/garage

• Same text encoder + VAE as T2V

TensorRT Engines (FL2V only):

• models/upscale_models/ → RealESRGAN_x4 (TensorRT engine)

• models/rife/ → rife425_ensemble_False_scale_1_sim (TensorRT engine)

TensorRT engines must be built for your specific GPU. See ComfyUI-RIFE-TensorRT-Auto and ComfyUI-Upscaler-TensorRT-Auto for build instructions.

Download Links

• Diffusion (FP8, T2V High): WAN REMIX T2V v2.0 High — Civitai

• Diffusion (FP8, T2V Low): WAN REMIX T2V v2.0 Low — Civitai

• Diffusion (FP8, I2V High): WAN REMIX v2.1 FP8 High — Civitai

• Diffusion (FP8, I2V Low): WAN REMIX v2.1 FP8 Low — Civitai

• Text encoder (NSFW FP8): nsfw_wan_umt5-xxl_fp8_scaled.safetensors

• Text encoder (standard FP8): umt5_xxl_fp8_e4m3fn_scaled.safetensors

• VAE: wan_2.1_vae.safetensors

🎬 WAN 2.2 Prompting Notes

How to Write Your Prompt

Describe the scene naturally. Be clear about the concepts below — Qwen3-VL handles the rest:

• 🎨 Visual style (put it first): photorealistic, cinematic, anime, 3D CG, claymation, vintage film, watercolor, fantasy

• 👥 Subjects: number, gender, appearance, clothing, position, expression

• 🏃 Action / motion: what happens, speed, interaction

• 🎥 Camera: dolly, pan, zoom, static, handheld, crane, orbit — smooth and continuous

• 🌍 Environment: setting, lighting, atmosphere, time of day

• 🔊 Audio (optional): connect MMAudio nodes to add synchronized sound

🔄 FL2V: Describe the transition between frames, not the scene (images fix the scene) 📖 Story: Write separate prompts for each timeline segment — Qwen3-VL handles the transitions

Resolution Guidance

WAN 2.2 native resolutions:

• 📱 Portrait: 832×1216 · 720×1280

• ⬛ Square: 1024×1024

• 🖥️ Landscape: 1216×832 · 1280×720

⚠️ Match the aspect ratio to your input image! Forcing 16:9 on a portrait image will squash it.

Duration

• Standard: 5 seconds (81 frames at 16 fps)

• Story/Timeline: up to 20 seconds (4 × 5s segments)

• Frame interpolation: RIFE doubles framerate to 48 fps where applicable

🎥 Camera Control Tags

All WAN 2.2 NSFW presets support camera control via the camera_tag dropdown on the QwenVL node — no need to type tags manually. Select from 19 camera movements:

• [STATIC_CAMERA] / [LOCKED_OFF] — Camera completely static

• [SLOW_ZOOM_IN] — Slow continuous push-in

• [SLOW_ZOOM_OUT] — Slow continuous pull-back

• [FAST_ZOOM_IN] — Fast aggressive push-in

• [FAST_ZOOM_OUT] — Fast pull-back, reveal context

• [PAN_LEFT] / [PAN_RIGHT] — Smooth horizontal pan

• [TILT_UP] / [TILT_DOWN] — Smooth vertical tilt

• [DOLLY_IN] / [DOLLY_OUT] — Physical dolly movement (parallax)

• [TRACKING_LEFT] / [TRACKING_RIGHT] — Lateral tracking shot

• [CRANE_UP] / [CRANE_DOWN] — Crane/jib movement

• [ORBIT] — Smooth 360-degree orbit around subject

• [HANDHELD] — Subtle handheld sway with micro-movements

• [ROLL] — Slow camera roll (rotation around lens axis)

How it works: the selected tag is injected at the start of the prompt AND as a FINAL CAMERA DIRECTIVE at the end, so Qwen respects it despite recency bias. The subject stays alive and active — the tag controls only the camera.

🎲 Wildcards

Selected workflows include a WildcardProcessor node that injects randomized prompt fragments from the PMP's Prompt Engine (__pmp/prmpt/*) wildcard library.

How It Works

1. The WildcardProcessor node sits before the Qwen3-VL prompt enhancer

2. At queue time, each __wildcard__ token is replaced with a random line from the corresponding .txt file

3. The expanded text is passed to Qwen3-VL, which converts it into the WAN 2.2 prompt format

4. Different seed = different wildcard picks — use a fixed seed for reproducible results

Customizing Wildcards

• Edit existing: open the .txt files under ComfyUI/custom_nodes/comfy-tagcomplete/wildcards/pmp/prmpt/

• Add your own: create a new .txt file, e.g. pmp/prmpt/mytags.txt, then reference it as __pmp/prmpt/mytags__

• Remove a wildcard: delete the __...__ token from the WildcardProcessor text field

• Disable randomization: replace the __wildcard__ token with a fixed string

Required Custom Node

• ComfyUI-TagComplete (includes the WildcardProcessor node and the __pmp/prmpt/* wildcard set) — huchukato/comfy-tagcomplete

The wildcard files ship with the custom node as fallback. On Docker/Vast.ai deployments, wildcards are downloaded from ComfyUI-Garage at boot for the latest version.

🐳 Docker / Cloud Ready

OneClick RunPod Template

Prefer a ready-to-go environment? Use the OneClick - ComfyUI - WAN 2.2 - Qwen3VL RunPod template:

• Docker image: huchukato/comfyui-qwenvl-runpod:cu13-wan22 (CUDA 13.0) or huchukato/comfyui-qwenvl-runpod:cu128-wan22 (CUDA 12.8)

• Base: huchukato/comfyui-base:cu130

• All custom nodes pre-installed

• ComfyUI Args: --disable-auto-launch --fast fp16_accumulation --use-sage-attention --cuda-malloc --async-offload

• All 8 workflows auto-downloaded at boot

• Models auto-downloaded at first boot (~62 GB including 4 WAN Remix diffusion models, NSFW text encoder, VAE; persistent)

• ComfyUI v0.34.2 baked into base image

• Sage Attention, FP16 accumulation, async offload

• TensorRT upscaling + RIFE interpolation

• PMP wildcards auto-downloaded from Garage at boot

Access: ComfyUI :8188 · JupyterLab :8888 · FileBrowser :8080 (user admin / password adminadmin12) · SSH ssh root@pod-ip

Vast.ai Provisioning

A Vast.ai provisioning script is also available:

• Script: vastai/wan22-provisioning.sh

• Downloads all models, workflows, wildcards, and custom nodes on first boot

• Same model set as RunPod Docker

ComfyUI Args (pre-configured)

--disable-auto-launch
--fast fp16_accumulation
--use-sage-attention
--cuda-malloc
--async-offload

🚀 Why Choose ComfyUI-QwenVL-Mod + WAN 2.2?

🎬 For Content Creators

• Multilingual: Write in any language, Qwen3-VL handles translation

• Story/Timeline: Multi-prompt timelines for long-form content (up to 20s)

• Quality: Native resolution, TensorRT upscale to higher resolution

🔥 For NSFW Content

• Explicit: Uncensored generation with dedicated NSFW presets

• Multiple presets: T2V, I2V (5s/20s), FL2V, Timeline — each tuned for its mode

• Detailed: Rich scene descriptions with explicit action

• Natural: Realistic progression, consistent characters

⚡ For Power Users

• Customizable: Easy to modify presets and system prompts

• Extendable: Add your own Qwen3-VL models (GGUF or HF)

• Optimized: Sage Attention, FP16, async offload, smart caching

• Multi-reference: image2 input for FL2V and SVI workflows

• Story: WanMoeKSampler + PainterI2V for complex multi-scene generation

🌟 What Makes This Special?

• Complete: 8 workflows covering T2V, I2V, FL2V, SVI, and Story

• Auto-prompting: Qwen3-VL handles prompt enhancement in any language

• Timeline: Multi-prompt Story workflows for up to 20-second videos

• TensorRT: Built-in upscaling and frame interpolation

• NSFW presets: Dedicated presets for each workflow type

• Wildcards: PMP prompt engine for randomized variation

• Docker-ready: OneClick RunPod template + Vast.ai provisioning

📋 Credits

• WAN 2.2 — Wan-AI · Comfy-Org/Wan_2.2

• ComfyUI — comfyanonymous/ComfyUI

• QwenVL-Mod — huchukato/ComfyUI-QwenVL-Mod

• Qwen3-VL — Qwen Team / Alibaba

• WAN Remix FP8 — FX_FeiHou

• NSFW Text Encoder — NSFW-API/NSFW-Wan-UMT5-XXL

• TensorRT RIFE / Upscaler — huchukato

• VideoHelperSuite — Kosinkadink

• Easy-Use — yolain

• PerfectVideoResolution — huchukato

• WanMoeKSampler — stduhpf

• PainterI2V / PainterLongVideo — princepainter

• mxToolkit — Smirnov75

• rgthree-comfy — rgthree

📄 License

Workflows are released under the same license as the underlying models and custom nodes. See each repository for details.

WAN 2.2 model weights: Wan-AI — Apache 2.0.

Built with ❤️ for the ComfyUI community