Details Overload | ComfyUI workflow — v1.2

This ComfyUI img2img workflow takes an input picture and uses Tile ControlNet, Detail Tweaker LoRA, and upscaling to add detail, producing a more detailed, upscaled image.

It combines Tile ControlNet, Detail Tweaker LoRA, and upscaling for detail levels described as subtle to absurd.

The listed nodes cover image loading, ControlNet application, VAE encoding and decoding, sampling, upscaling, and image saving.

The image path includes LoadImage for the picture, ImageUpscaleWithModel for upscaling, and SaveImage for saving the result.

Install WAS Node Suite, place Detail Tweaker LoRA in your LoRA folder and the ControlNet Tile (e) model in your ControlNet folder, then download and open the workflow.

Version v1.2 is described as fixing a bug caused by a WAS Node Suite update in March.

The named model files are 4x_NMKD-YandereNeoXL_200k.pth, add_detail.safetensors, controlnet11Models_tileE.safetensors, onlyanime_v2-840000.safetensors, and vae-ft-mse-840000-ema-pruned.vae.pt.

The workflow lists CheckpointLoaderSimple, CLIPSetLastLayer, CLIPTextEncode, Constant Number, ControlNetApplyAdvanced, ControlNetLoader, Debug Number to Console, ImageScaleBy, ImageUpscaleWithModel, KSampler, Latent Input Switch, LatentUpscaleBy, LoadImage, and LoraLoader.

It also lists Number Input Switch, Number Operation, Number to Float, Number to Text, SaveImage, Text Contains, Text Input Switch, Text Multiline, Text String, Text to Console, UpscaleModelLoader, VAEDecode, VAEEncode, and VAELoader.

All parameters are described as tuned, with one metric to change; the metric is not named.

The listed setup uses SD 1.5 with OnlyAnime, vae-ft-mse-840000-ema-pruned, 4xNMKD YandereNeo XL, EasyNegative, and badhandv4.

The listed substitutions are any anime or semi-realistic model, 4xUltraSharp for the upscaler, or no textual inversions.

No model-use license is specified.