Sign In

SD 1.5 Hyper workflow, Simple and semi-advanced versions (Settings for realistic images)

SD 1.5 Hyper workflow, Simple and semi-advanced versions (Settings for realistic images)

Just wanted to post the current settings I am using for using the wonderful new LORA created by ByteDance which enables sd 1.5 checkpoints to make quality images in 8 steps which unlike LCM retains cfg for negative prompting.
https://civitai.com/models/428900?modelVersionId=477885

I've made 2 versions, one which has a bit more than the one posted by ByteDance and a semi-advanced version with upscale and FreeU.

My versions are setup for PicX_real but should work just fine with other realistic checkpoints
https://civitai.com/models/241415?modelVersionId=272376

For the "advanced" version there are multiple places where you can change stuff... (I dont know if you can disable widgets somehow?) so here's where you set stuff:

  • Seed: SamplerCustom

  • Height/Width: CR SD1.5 Aspect Ratio

  • Batch size: pipeLoader

Things that you can tweak is:

  • TCD Model Sampling Discrete: ETA

  • FreeU_V2: All the settings

  • SamplerCustom: CFG

Use the keybindings 1,2,3,4,5 to move your view around.

Enjoy :)

You can find the "original" (most barebone) workflows here:
https://huggingface.co/ByteDance/Hyper-SD/tree/main/comfyui

6

Comments