Nodes/ComfyUI-GGUF-Loader/Krea2 img2img ⚡
ComfyUI Node

Krea2 img2img ⚡

Prompts, init latent and control-image attach for Krea2. Leave image unconnected for txt2img. Feed the outputs straight into a stock KSampler.

By ChrisColeTech·Created 13 days ago·Updated about 3 hours ago· 6
Krea2 img2img ⚡
  • model
  • clip
  • vae
  • image
  • control_image
  • edit_reference
  • model
  • positive
  • negative
  • latent
  • denoise
prompt
negative_prompt
strength0.60
batch_size1
width1024
height1024
control_modeauto_depth
depth_ckpt_namedepth_anything_v2_vitb.pth
control_channel_modegrayscale
control_normalizeper_image_minmax
control_invertfalse
control_batch_modeindependent_images
Category🤖 CCTech/Krea2

Inputs (18)

NameTypeDefaultDescription
modelMODEL
clipCLIP
vaeVAE
promptSTRING
negative_promptSTRING
strengthFLOAT0.600–1img2img only. How much of the init image to discard. Ignored without an image.
batch_sizeINT11–4096
widthINT102416–16384Output size. With an init or control image this resizes it.
heightINT102416–16384
imageoptIMAGEInit image for img2img, and (in auto_depth/auto_canny modes) the source photo the control image is derived from. Leave unconnected for txt2img.
control_modeoptCOMBOauto_depthauto_depth/auto_canny: derive the control image from `image` automatically - pick whichever matches the loaded Control LoRA. manual: no automatic derivation, connect control_image yourself - use this for any Control LoRA the two auto modes don't cover (pose/lineart/normal).
depth_ckpt_nameoptCOMBOdepth_anything_v2_vitb.pthauto_depth mode only. Model size for the automatic depth estimation. Downloads on first use if not already in models/depth_anything_v2.
control_imageoptIMAGEManual control map - a depth/canny/pose/etc. map. Overrides auto_depth when connected. Required in manual mode.
control_channel_modeoptCOMBOgrayscalegrayscale for depth; rgb for canny/pose/lineart/normal.
control_normalizeoptCOMBOper_image_minmaxper_image_minmax for depth; none for canny/pose/lineart/normal.
control_invertoptBOOLEANfalseFlip if the preprocessor's convention is reversed from the LoRA's training convention (e.g. depth preview shows near objects dark instead of white).
control_batch_modeoptCOMBOindependent_images2 options: independent_images, video_frames
edit_referenceoptIMAGEFor an in-context/edit-style Krea2 LoRA (e.g. krea2_canny-v0.1.safetensors - an ordinary LoRA, loaded via stock LoraLoaderModelOnly, NOT Krea2ControlLoRALoader): the reference image (a canny map, for that LoRA). Encoded and attached to positive conditioning as reference_latents - separate from img2img's `image` and the widened-projection control_image above.

Outputs (5)

NameTypeDescription
modelMODEL
positiveCONDITIONING
negativeCONDITIONING
latentLATENT
denoiseFLOAT