OpenArt
OpenArt
Workflows

    My stuff

  • My Workflows

  • Liked Workflows

  • Following Workflows

Go to OpenArt main site
Upload workflow

Blake's Stable Diffusion 1.5 Overkill Upscale Workflow

5.0

0 reviews
12
4.6K
1.3K
1
Description

Create and upscale images to 5120 x 8192 (or whatever resolution) using Stable Diffusion 1.5, Perturbed-Attention Guidance, Perp-Neg, Kohya High Res Fix, ControlNet, CLIPSeg Face Detailer, and more.

7/8/2024

  • Another massive update. I remade most of the stages. I now use Perp-Neg for stage 1 and 2. This made a big improvement.
  • I added Perturbed-Attention Guidance.
  • I added a LoRa stack.
  • I changed to using the regular Mask Detailer, since the previous issues have been fixed. I ran into issues with close up images, using the previous more custom detailer.
  • Cleaned up the ControlNet Upscaling.
  • Removed all the switches, since I don't use them in my workflow.

2/1/2024

  • Massive update with what I learned from SDXL workflow. Everything is now locked down. Added save as jpg at the end.
  • Added a custom CLIPSeg detailer for faces, or whatever you want. I didn't like the default face detailers, this is MUCH faster and works better. It removes/masks the face, uses an upscaler 4x, down-samples to 2x, masks the face again (at more detail), details the face, down-samples again to 1x, and then puts the face back in place. Super easy, barely an inconvenience!
  • You can use any resolution now and set the upscale ratio for each stage. It also now displays the resolution for each Stage.
Discussion

(No comments yet)

Loading...

Author

Blake

Blake

No reviews yet

  • - latest (a year ago)

  • - v20240202-042318

  • - v20240118-030323

Primitive Nodes (0)

Custom Nodes (97)

Comfyroll Studio

  • - CR Integer To String (4)

  • - CR LoRA Stack (1)

  • - CR Apply LoRA Stack (1)

ComfyUI

  • - ImageBlend (4)

  • - ImageScale (3)

  • - ImageUpscaleWithModel (4)

  • - VAEDecode (3)

  • - CLIPTextEncode (3)

  • - EmptyLatentImage (1)

  • - ControlNetApplyAdvanced (2)

  • - VAEEncode (2)

  • - ConditioningCombine (2)

  • - ControlNetLoader (2)

  • - UpscaleModelLoader (4)

  • - PreviewImage (12)

  • - BasicScheduler (2)

  • - KSamplerSelect (2)

  • - PerpNegGuider (2)

  • - CheckpointLoaderSimple (1)

  • - MaskToImage (1)

  • - SamplerCustomAdvanced (2)

  • - KSampler (1)

  • - PatchModelAddDownscale (1)

  • - RandomNoise (2)

ComfyUI Impact Pack

  • - ImpactFloat (2)

  • - ToBinaryMask (1)

  • - ToBasicPipe (1)

  • - MaskDetailerPipe (1)

ComfyUI Nodes for Inference.Core

  • - LeReS-DepthMapPreprocessor (1)

  • - MiDaS-DepthMapPreprocessor (1)

  • - Zoe-DepthMapPreprocessor (1)

  • - OpenposePreprocessor (1)

Perturbed-Attention Guidance

  • - PerturbedAttention (1)

pythongosssss/ComfyUI-Custom-Scripts

  • - MathExpression|pysssss (4)

  • - ShowText|pysssss (4)

Save Image with Generation Metadata

  • - Width/Height Literal (2)

  • - Sampler Selector (1)

  • - Cfg Literal (1)

  • - Seed Generator (2)

  • - Scheduler Selector (1)

  • - Int Literal (1)

  • - Checkpoint Selector (1)

  • - String Literal (3)

WAS Node Suite

  • - Image Save (3)

  • - CLIPSeg Masking (1)

  • - Mask Gaussian Region (1)

Checkpoints (2)

0.3(0.5(arthemyComics_v40) + 0.5(flat2DAnimerge_v30)) + 0.7(0.5(analogDiffusion_10Safetensors) + 0.5(realisticVisionV20_20)).ckpt

0.5(realcartoonRealistic_v7) + 0.5(0.7(0.7(0.5(dreamshaper_7) + 0.5(animesh_PrunedV21)) + 0.3(disneyPixarCartoon_v10)) + 0.3(0.5(arthemyComics_v40) + 0.5(flat2DAnimerge_v30))).ckpt

LoRAs (2)

civitai logo

add_detail.safetensors

epiNoiseoffset_v2.pt