skills.rest
sota411
sota411COMMUNITY
Explore all skills by this author →
GitHub

wan22-kijai-video

Prepare, run, resume, and diagnose Wan2.2 I2V video generation with Kijai WanVideoWrapper and LightX2V LoRA.

1|Updated Jul 16, 2026
Search Tags:#video generation#lora#i2v#comfyui#wan2.2#kijai#rtx 3080
One-click install
npx skills add https://github.com/sota411/codex-config --skill wan22-kijai-video-sota411
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: wan22-kijai-video
Source: https://github.com/sota411/codex-config/tree/main/user-skills/wan22-kijai-video
Command: npx skills add https://github.com/sota411/codex-config --skill wan22-kijai-video-sota411

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Running Wan2.2 image-to-video generation on a 10 GB RTX 3080 through ComfyUI involves fragile model pins, LoRA SHA verification, queue conflicts, and multi-beat workflows where a single mistake wastes hours of GPU time. This Skill enforces a safe, repeatable procedure for preparing, executing, resuming, and diagnosing Kijai WanVideoWrapper jobs. ## Core Features & Use Cases - Preflight validation: Verifies WanVideoWrapper commit, high/low LightX2V 4-step LoRA SHA-256 hashes, template SHAs, and the WanVideoConditioningLimit node before any generation, failing fast on mismatches. - Safe queue discipline: Reads the ComfyUI /queue before submitting, never interrupts other users' prompts, never calls the global /free endpoint, and only restarts when idle after node changes. - Beat-gated production workflow: Runs one 21-frame beat at a time via run-beat, requires human motion audit and explicit approval of the preview profile before starting the 1280x720 production profile, and resumes safely via journal/history reconciliation. - Use Case: You want to generate a 4-beat coastal stairway video on your home RTX 3080. The Skill dry-runs the pinned config, validates LoRA hashes, executes beat 1 in preview, waits for your motion audit approval, then proceeds beat by beat while logging journal and prompt IDs. ## Quick Start Use the wan22-kijai-video skill to dry-run and then run beat 1 of the coastal-stairway preview profile on my RTX 3080 ComfyUI setup.

Frequently Asked Questions about wan22-kijai-video

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run Wan2.2 I2V video generation with Kijai WanVideoWrapper?▼

Run the pinned config through the runner: first a dry-run, then run-beat --beat N for a single 21-frame beat. The workflow uses high/low models at 2 steps each (4 total), CFG 1, Euler sampler, and the LightX2V sigma schedule, with SHA verification before submission.

What settings work for Wan2.2 video generation on an RTX 3080 10GB?▼

Use WanVideoWrapper FP8 scaled high/low models with 22-block swap, CPU/offload device, and SDPA attention. Do not enable torch compile, SageAttention, or TeaCache without a separately approved benchmark, and do not substitute Kaggle Q4 GGUF models.

Can I use Kaggle to generate Wan2.2 videos with this workflow?▼

No. The Kaggle artifact is a pinned preflight only for verifying models and Kijai nodes; it does not submit video jobs. It requires a private wan22-i2v-models input with the text encoder, FP8 models, LoRAs, and VAE at documented paths.

Why does long-prompt encoding fail in WanVideoWrapper workflows?▼

Native T5 embeddings can exceed token limits, so prompts must be routed through CLIPTextEncode into WanVideoConditioningLimit with max_tokens=512, then WanVideoTextEmbedBridge. Manually truncating prompt strings to bypass this is not allowed.

How do I safely resume an interrupted ComfyUI video generation job?▼

Reconcile the run journal against ComfyUI history to determine the submission outcome before doing anything. If the outcome is unknown, do not resend with --resume; never restart or submit while /queue shows another running or pending prompt.

Related Skills

Explore More Design & Creative Skills→
fumiya-kumefumiya-kume
1

Wan2.1/2.2 I2V ComfyUI Implementation

Implement Wan2 I2V workflows in ComfyUI with GGUF configurations.

Community
Advanced
theSamPadillatheSamPadilla
25

write-overlay

Generate frame-driven JSX overlay components for Montaj video compositing.

Community
Intermediate
Supreme-UltimateSupreme-Ultimate
146

animator-skill

Convert static storyboard frames into concise motion prompts for AI video generation.

Official
Intermediate
freedomw1987freedomw1987
2

comfyui

Deploy, configure, and execute ComfyUI generative media workflows locally or in the cloud.

Community
Advanced