## GingerLabs Wan 2.2 T2V A14B

A GingerLabs-tuned derivative of [Wan-AI/Wan2.2-T2V-A14B](https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B), packaged as matching HighNoise and LowNoise diffusion-model pairs for ComfyUI.

Available formats:

- GGUF Q4_K_M

- GGUF Q5_K_M — recommended balance

- GGUF Q6_K

- GGUF Q8_0

- Scaled FP8 E4M3FN

No acceleration LoRA is required.

### Important: download both files

Each version requires its matching HighNoise and LowNoise files. Keep both files at the same quantization level and do not swap their roles.

Example:

- GingerLabs_Wan2.2_T2V_A14B_High_Q5_K_M.gguf

- GingerLabs_Wan2.2_T2V_A14B_Low_Q5_K_M.gguf

Place both files in:

ComfyUI/models/diffusion_models/

### Requirements

- [ComfyUI-WanVideoWrapper](https://github.com/kijai/ComfyUI-WanVideoWrapper) 1.3.6 or newer

- ComfyUI-VideoHelperSuite

- Wan2_1_VAE_bf16.safetensors

- Exactly nsfw_wan_umt5-xxl_bf16.safetensors from [NSFW-API/NSFW-Wan-UMT5-XXL](https://huggingface.co/NSFW-API/NSFW-Wan-UMT5-XXL/tree/main?not-for-all-audiences=true)

The scaled-FP8 version of that UMT5 encoder is not supported by the included workflow.

Place the supporting files in:

- Text encoder: ComfyUI/models/text_encoders/

- VAE: ComfyUI/models/vae/

### Recommended starting settings

- Total steps: 8

- HighNoise range: 0–4

- LowNoise range: 4–end

- Sampler: Euler

- Shift: 5.0

- CFG: 1.0

- Resolution: 720 × 1280

- Frames: 33

- FPS: 16

This means 8 total denoising steps, split as four HighNoise and four LowNoise steps—not 8+8.

At CFG 1.0, negative conditioning is skipped, so the negative prompt does not affect sampling.

For the WanVideoWrapper model loaders, leave quantization set to disabled. The wrapper loads GGUF directly and detects the scaled-FP8 format. SDPA is the portable attention default; SageAttention is optional.

### Scope

- Text-to-video only

- Not intended for I2V, TI2V, FLF2V, S2V, Animate, or VACE

- No trigger words

- Both HighNoise and LowNoise files are required

- Frame counts should follow 4n+1

The files passed structural and load/unload validation. A comparative visual-quality or performance benchmark between quantization levels has not been performed.

### Downloads and workflow

- [Reference ComfyUI workflow](https://huggingface.co/GingerLabsPlatform/GingerLabs-Wan2.2-T2V-A14B/resolve/main/workflows/GingerLabs_T2V_Workflow.json)

- [Full model card, files, and checksums](https://huggingface.co/GingerLabsPlatform/GingerLabs-Wan2.2-T2V-A14B)