Wan 2.2 Detailer — v1.0

Author notes from Wan 2.2 Detailer - DirkDigitaller.json · selected release

### Wan NAG - will highlight the most important prompt elements for the Wan model to focus on. Not necessary, but very useful. ### Reactor FaceSwap - will attempt to give character consistency to the finished image. This works in one of 2 ways; you can either use the original reference image and the workflow will attempt to crop the face from the original image, or you can upload a clear image with the character desired for faceswap into the 'FaceSwap Image' node. If the original image has multiple people, there is a chance the workflow will grab the wrong face - in this case, you should activate the 'FaceSwap Image' node. ### FaceSwap Image - see above. Only activate if Reactor FaceSwap is activated and you want to upload an image for faceswap. If this node is not active but Reactor FaceSwap is active, Reactor will attempt to use the Qwen generated image for faceswap. If you upload to this node, make sure the image uploaded has a clear focus on your subject, preferrably with no other faces in frame.

Author notes from Wan 2.2 Detailer - DirkDigitaller.json · selected release

### [Wan 2.2 T2V Low Noise FP8 Scaled](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors?download=true) ### [Wan 2.1 VAE](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors) ### [Wan 2.2 CLIP Text Encoder](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors) ### [Wan 2.2 lightx2v 4-Step LoRA](https://huggingface.co/lightx2v/Wan2.2-Lightning/resolve/main/Wan2.2-T2V-A14B-4steps-lora-rank64-Seko-V2.0/low_noise_model.safetensors?download=true) ### Recommended Custom Node - Quick Connections

Author notes from Wan 2.2 Detailer - DirkDigitaller.json · selected release

## Multi Model T2I Workflow (Qwen ImageEdit 2511 + Wan 2.2 T2V) ### [ DirkDigitaller @ CivitAI](https://civitai.com/user/DirkDigitaller)

Author notes from Wan 2.2 Detailer - DirkDigitaller.json · selected release

## Ways To Modify Consistency/Divergence: ### Reactor FaceSwap - will attempt to create facial consistency between images. This can be turned on/off from 'Options'. The settings of the Reactor FaceSwap nodes can increase facial consistency but may also reduce emotional expression consistency. Try adjusting 'face_restore_visibility' (0.1 to 1) and 'codeformer_weight' (0.1 to 1) to get the balance between facial consistency and emotional expression consistency. ### ClownsharkKSampler -> eta - this parameter changes the noise added during the sampling process. Range from 0.1 up to 2, with 0.1 having high consistency (use for detailing) and 2 having high divergence (use for creative styling). ### ClownsharkKSampler -> steps - adjusting the total steps in this node will also impact the amount of work Wan does - higher step count will create higher divergence but increase potential for over-cooking, lower step count will create higher consistency but increase potential for undercooking.