Skip to content
WorkflowListedExternal
Open source

Wan2.2 I2V_T2I_14B_gguf — T2I_I2V_V1

If you find the workflow isn't working right, go ahead and redownload.

Sourcecivitai.com/models/1820099
TaskMultiCapabilitiesImage to image · UpscalingFamilyWanEstimated VRAMNot yet characterizedBgrade

Runtime profile

Access
External workflow
Status
Listed
Source
Civitai
Updated
Oct 2, 2026

Source description

Update-20250730:

If you find the workflow isn't working right, go ahead and redownload.

Just finished my first two Wan2.2 LoRAs today—one trained on the high-noise checkpoint, the other on the low-noise— so I’ve updated the workflow for a couple of reasons:

Turns out the Low-noise checkpoint is the real final boss. Some people have already spotted that if you only call your character LoRA during the high-noise stage, the Low-noise pass will basically torch everything you just did.

Think of it this way:

A = LoRA trained only on the high-noise ckpt

B = LoRA trained only on the low-noise ckpt

A+B = both combined

Quality comes out in this order: A+B > B > A

The uploaded zip file contains two workflows: I2V and T2I.

They're tweaked versions of the ComfyUI workflow templates - just minor adjustments, arranged the way I like, and with modified generation parameters (I put my test results in my daily update article).

Also, go with the flow, I added upscaling and noise nodes to T2I. They're simple nodes, not using a repaint upscaling method like A1111, to make sure they can run on my PC.

What you need

Model files
16.3 GiB known subtotal

Remaining hardware or download-size estimates are not available yet.

Requirements

Models

7

Node packs

3

  • ComfyUI-GGUF

    Possible

    Node pack · Registry

  • comfyui-videohelpersuite

    Found

    Node pack · Registry

  • rgthree-comfy

    Found

    Node pack · Registry

AI-generated commentary

AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.

Purpose

  • Use Wan2.2 I2V_T2I_14B_gguf in ComfyUI to run separate I2V and T2I generation workflows; the description names these modes but does not specify concrete input or output types.

Flow

  • The zip contains two workflows: I2V and T2I.

  • T2I includes simple upscaling and noise nodes.

Inputs and outputs

  • No concrete input or output types are specified.

Setup

  • Published by SamLiu, these are lightly adjusted ComfyUI templates based on Wan Video 14B i2v 480p.

Dependencies

  • The workflow names these ComfyUI node packs: ComfyUI-GGUF, comfyui-videohelpersuite, and rgthree-comfy.

  • Files include 4xLexicaHAT.pth, clip_vision_h.safetensors, umt5_xxl_fp8_e4m3fn_scaled.safetensors, and wan_2.1_vae.safetensors.

  • Other files include wan2.2_i2v_high_noise_14B_Q4_K_S.gguf, wan2.2_i2v_low_noise_14B_Q4_K_S.gguf, and Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64.safetensors.

Configuration

  • The workflows use modified generation parameters.

  • T2I uses simple upscaling and noise nodes rather than a repaint upscaling method like A1111.

  • For character LoRAs, A is high-noise-only, B is low-noise-only, and A+B combines both; the stated order is A+B > B > A.

Cautions

  • Before use, verify that ComfyUI-GGUF, 4xLexicaHAT.pth, and Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64.safetensors are available under those exact names.

    Suggestion · not verified

  • If the workflow is not working right, redownload it.

    Suggestion · not verified

Does this need to be edited?

Sign in to send an edit request.

Sources

1 source
CivitaiPrimary
civitai.com/models/1820099

Verified Jul 19

Source excerpts

1 excerpt
NoteJul 19civitai.com/models/1820099

Source context: 1204 downloads · Type Workflows · Base model Wan Video 14B i2v 480p