https://limewire.com/d/xkuSy#ttzf4p0aSx

mrdion Much appreciated, workflow is working well with this lora 😘

blacklabe188 it's a basic workflow you can find in comfyui template

qazxsw Thanks, i am new and trying harder to learning comfyui how to use 😁

mrdion sorry im pretty new to flux and confyui i loaded the workflow and i selected the image i want to modify, how do i place the object i want to insert on the top of the img? do i just modify it in photoshop? im confused can you explain it to me?

Alefx69 use win11 paint app , place the APPLE to the girl mouth area" [save it], then selected this image to this workflow + lora and [RUN]

blacklabe188 ok thanks i will try again!

mrdion can you reupload the workflow, please?

alyssaposta https://limewire.com/d/ttJ1O#wUOgogq41L

mrdion thank you

the error is the message. You need to add a dummy weight as the last weight.

obanasiak601 in nunchaku lora loader node there is only one value you can change called "lora_strenght"

Place it hat (is not working, comfyui Generates very nice hair to replace the hat 😂)

Yeah, same problem.

Yeah, same problem.

Yeah, same problem.

Same here, so sad theres no enough info

cldoerick131 "place it, your object" works great. whatever you cover, kontext will make up hence why if you cover the face it will change the face cuz it literally does not see what is behind it

same

Me too. It seems to be a nunchaku issue.

i can't get it to work . Is the " ," important ? it's "place it, my object" or "place it my object" exactly ?

like do we just input one image with teh face over another face and go? or do you need a workflow with 2 separate image inputs?

@lionelelian731 if i was trying to place a hat, and below was the prompt node i would do this:

place it, the hat

i tried editing my comment because i see it can be confusing but it wont work.

or nunchaku will patch support for it. there's an open issue already at least

tazztone nunchaku team is already working on too many things including supporting VACE and other stuff, if its possible for Place it author to patch it! It'd be 100% easier and we'll get faster support like within a couple of weeks, but asking nunchaku team to do it is like we'll have to wait at least 2~3 months to get it because their hands is already full!

Ryudaio Ryudaio true. u can patch the lora with this https://github.com/tazztone/patch_comfyui_nunchaku_lora

worked well for me

you failed

What prompt did you use for swapping face? I used "place it face" but didn't get good result.

skymidrevengers873 I am not exactly sure right now. Give "Place it head" also a try. Make sure that the head you place on is roughly the same size as the original head. Only swapping the face might not be good enough or even possible. I created a lora of a person that gives me good enough results for the head, hair and body. After generating that image, I use place it and a real photo to replace the whole head, resulting in a perfect face of that person.

fenasikerim bro şunu türkçede açıklar mısın ya? photoshop ile kesilmiş fotoğrafı yerleştiriyorum place it face veya head dedim ama hafif benzeyen yüz oluyor şekli tam oturmamış,tam oturtamıyor

gokaygokay anladım teşekkürler bide workflow kaynaklı mı bilmiyorum 2 fotoğraflı resimleri zor birleştiriyor,prompt'u uygulamıyor çoğu zaman.

gokaygokay bilgilendirme için teşekkürler şimdi 6000x8000 çözünürlükte tek photo olarak denedim deniyordum zaten, mesela "place it jersey" forma yerleştirmek istedim fakat işe yaramadı,kullandığın workflow paylaşabilir misin ? workflow kaynaklı olduğunu düşünüyorum.

mthnzz Hacı zaten çok fazla benzeyen resim çıkmıyor ortaya. Ben Place it ile kafayı yerleştirip, sonradan tekrar yüze face swap yapıyorum lora ve referans resimler kullanarak.

the nod fails to load in comfyui startup. neither is it in repository

testingbetas7604 Hi, thanks for the report! This loading issue should be fixed now. Please try reinstall it.

Same issue here. Doesn't show up in the nodes' list

PASTE INTO BAT FILE: 0<0# : ^

'''

@echo off

set script=%~f0

python -x "%script%" %*

exit /b 0

'''

import os

import torch

import safetensors.torch

from safetensors import safe_open

import tkinter as tk

from tkinter import ttk, messagebox

from tkinterdnd2 import DND_FILES, TkinterDnD

def patch_final_layer_adaLN(state_dict, prefix="lora_unet_final_layer"):

final_layer_linear_down = None

final_layer_linear_up = None

adaLN_down_key = f"{prefix}_adaLN_modulation_1.lora_down.weight"

adaLN_up_key = f"{prefix}_adaLN_modulation_1.lora_up.weight"

linear_down_key = f"{prefix}_linear.lora_down.weight"

linear_up_key = f"{prefix}_linear.lora_up.weight"

if linear_down_key in state_dict:

final_layer_linear_down = state_dict[linear_down_key]

if linear_up_key in state_dict:

final_layer_linear_up = state_dict[linear_up_key]

has_adaLN = adaLN_down_key in state_dict and adaLN_up_key in state_dict

has_linear = final_layer_linear_down is not None and final_layer_linear_up is not None

if has_linear and not has_adaLN:

dummy_down = torch.zeros_like(final_layer_linear_down)

dummy_up = torch.zeros_like(final_layer_linear_up)

state_dict[adaLN_down_key] = dummy_down

state_dict[adaLN_up_key] = dummy_up

return True # Was patched

return False # Nothing changed

def patch_file(input_path, output_folder, status_callback):

try:

state_dict = {}

with safe_open(input_path, framework="pt", device="cpu") as f:

for k in f.keys():

state_dict[k] = f.get_tensor(k)

patched = False

prefixes = [

"lora_unet_final_layer",

"final_layer",

"base_model.model.final_layer"

]

for prefix in prefixes:

before = len(state_dict)

did_patch = patch_final_layer_adaLN(state_dict, prefix=prefix)

after = len(state_dict)

if did_patch and after > before:

patched = True

break

filename = os.path.basename(input_path)

if patched:

out_path = os.path.join(output_folder, filename)

safetensors.torch.save_file(state_dict, out_path)

status_callback(f"✅ Patched and saved: {filename}")

else:

status_callback(f"⚠️ Skipped (already patched or missing final_layer): {filename}")

except Exception as e:

status_callback(f"❌ Error patching {os.path.basename(input_path)}: {e}")

class App(TkinterDnD.Tk):

def init(self):

super().__init__()

self.title("LoRA adaLN Patcher - Drag & Drop")

self.geometry("600x400")

self.output_folder = os.path.join(os.getcwd(), "patched_outputs")

os.makedirs(self.output_folder, exist_ok=True)

lbl = ttk.Label(self, text="Drag and drop your .safetensors files here", font=("Arial", 14))

lbl.pack(pady=10)

self.drop_area = tk.Text(self, height=15, width=70)

self.drop_area.pack(padx=10, pady=10)

self.drop_area.insert("1.0", f"Output folder:\n{self.output_folder}\n\nDrop files below:\n")

self.drop_area.config(state=tk.DISABLED)

self.drop_area.drop_target_register(DND_FILES)

self.drop_area.dnd_bind('<<Drop>>', self.on_drop_files)

def on_drop_files(self, event):

files = self.split_filenames(event.data)

for f in files:

if f.endswith(".safetensors"):

self.log_status(f"Patching file: {os.path.basename(f)}")

patch_file(f, self.output_folder, self.log_status)

else:

self.log_status(f"Skipped (not a .safetensors file): {os.path.basename(f)}")

def split_filenames(self, file_string):

# Handles splitting multiple files depending on OS drag & drop format

if self.tk.call('tk', 'windowingsystem') == 'aqua':

# macOS wraps paths in {}

import re

return re.findall(r'{([^}]+)}', file_string)

else:

return self.tk.splitlist(file_string)

def log_status(self, message):

self.drop_area.config(state=tk.NORMAL)

self.drop_area.insert(tk.END, message + "\n")

self.drop_area.see(tk.END)

self.drop_area.config(state=tk.DISABLED)

if name == "__main__":

app = App()

app.mainloop()

Nunchaku-Flux-Kontext workflow - https://civitai.com/posts/25403381

@xNzX Way too outdated. Half of the shit is broken. I mean try to run it on fresh latest Python 3.12 ComfyUI portable, the cu128 version afaik.
Basically the insightface doesn't work as usual so I switched to dlib in the FaceAnalysis node with the .dat models installed, then it says "Unsupported image type, must be 8bit gray or RGB image." which they are and apparently this issue is because of numpy but after installing 1.26.4 the KJNodes breaks...

@exacomedia670 I don't use KJNodes and not knowning about them. My Python is 3.12, pytorch version: 2.8.0+cu126, numpy 1.26.4 — all works fine.

@xNzX Without KJNodes the Get/Set nodes are red for me. Will try the same versions of yours, thanks.

@exacomedia670 "KJNodes the Get/Set nodes" — Ah, OK, I remember them. It works for me. I just not use it - and remove it from my ComfyUI.

hey, you can convert it using rbg lora converter. i did it and it worked like a charm.

Nunchaku-Flux-Kontext workflow - https://civitai.com/posts/25403381