Using Comfyui

[Bagel94]

So, I have started the process of getting comfyUI installed and have started to use it.

I have everything installed, and all the recommended model configurations, but all I keep getting is a white blob. Any idea as to what is wrong?

[haico1992]

I don't even get to this part, received
AI replacement failed: AI generation failed: 405: Method not allowed
RapidRaw 1.5.0 , ComfyUI 0.15.0
Is there anything else I should install in ComfyUI?

[cogitech2]

Same problem as haico1992 here. Nothing in the logs of RapidRAW nor ComfyUI to indicate what the issue is.

The ComfyUI integration is something that really needs more documentation, or maybe I just can't find it? My ComfyUI works perfectly via the web GUI and the RapidRAW connection test works as well.

Bagel94 said "I have everything installed, and all the recommended model configurations" which makes me think I am missing something very important.

[cogitech2]

I made quite a lot of progress on this today. I discovered that "RapidRAW AI Connector" is required, so I installed and configured that. Then I grep'd the connector's workflow.json to find what model files (.safetensors) I needed. Then I used google AI to help me find those. Then I made sure the filenames in the workflow.json matched perfectly with the actual .safetensors files that I downloaded (into the correct directories under "/models/"). Then I configured RapidRAW to connect to the connector (127.0.0.1:5000) and tried it out. There was an error regarding a missing parameter "device_mode" so I used google AI to troubleshoot and it told me that the "InpaintCropImproved" node has a new parameter to choose between CPU and GPU. I had to add "device_mode": "gpu (much faster)", to the workflow.json file in the section full of parameters under ""title": "✂️ Inpaint Stitch (Improved)" "

Now it works! Results are mixed (as is typical with generative AI) but pretty quick (RTX 3060 12GB) and fun to play with.

[haico1992]

No way people would figure it out on their own, including me. Thank!

…

On Wed, 4 Mar 2026, 08:27 cogitech2, ***@***.***> wrote: I made quite a lot of progress on this today. I discovered that "RapidRAW AI Connector" is required, so I installed and configured that. Then I grep'd the connector's workflow.json to find what model files (.safetensors) I needed. Then I used google AI to help me find those. Then I made sure the filenames in the workflow.json matched perfectly with the actual .safetensors files that I downloaded (into the correct directories under "/models/"). Then I configured RapidRAW to connect to the connector ( 127.0.0.1:5000) and tried it out. There was an error regarding a missing parameter "device_mode" so I used google AI to troubleshoot and it told me that the "InpaintCropImproved" node has a new parameter to choose between CPU and GPU. I had to add "device_mode": "gpu (much faster)", to the workflow.json file in the section full of parameters under ""title": "✂️ Inpaint Stitch (Improved)" " Now it works! Results are mixed (as is typical with generative AI) but pretty quick (RTX 3060 12GB) and fun to play with. — Reply to this email directly, view it on GitHub <#547 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACFE2YTL2RMXTBPEAZKJXJD4O6BBDAVCNFSM6AAAAACQGBNEQSVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKOJZGE2DEMY> . You are receiving this because you commented.Message ID: ***@***.***>

[cogitech2]

You're welcome. I'm not smart but I don't give up.

[cogitech2]

realvisxlV50_v50LightningBakedvae.safetensors ------> /your/path/to/comfyui/models/checkpoints

(above is the closest model I could find to the one mentioned in the workflow.json)

diffusion_pytorch_model_promax.safetensors -------> /your/path/to/comfyui/models/controlnet

sdxl_vae.safetensors ---------> /your/path/to/comfyui/models/vae

[haico1992]

where is the workflow.json file that you have edit located?

[cogitech2]

After you follow the installation instructions here : https://github.com/CyberTimon/RapidRAW-AI-Connector you will find the workflow.json in the "RapidRAW-AI-Connector" directory.

By the way, I highly recommend that you create a python virtual environment to run this in before installing it. For example, here's what I did:

---

git clone https://github.com/CyberTimon/RapidRAW-AI-Connector.git
cd RapidRAW-AI-Connector/
python3 -m venv RR-connector
source RR-connector/bin/activate
pip install -r requirements.txt

After this, edit the workflow.json as i described earlier and get the right safetensors files in the right directories.

Then try running it:

COMFY_HOST=127.0.0.1 COMFY_PORT=8188 python main.py

(assuming your comfyUI is running on localhost port 8188 -- change accordingly if necessary)

Then point RapidRAW application to use 127.0.0.1:5000 for AI (the connector is listening here)

[cogitech2]

I should have mentioned I am running all this on Debian Linux, so it may or may not work depending on your Linux distro (or other OS?)

[GimleLarpes]

I threw together a short guide on how to get everything up and running, hope it helps someone. The guide is based on Linux / bash, however the command translation is relatively straight forward.

Installing ComfyUI

To install ComfyUI, follow the instructions on the wiki.

Installing models

To download the models used in the default connector workflow.json: (execute in your ComfyUI directory)

# Download models
comfy model download --url https://huggingface.co/stabilityai/sdxl-vae/resolve/main/sdxl_vae.safetensors --relative-path models/vae
comfy model download --url https://huggingface.co/xinsir/controlnet-union-sdxl-1.0/resolve/main/diffusion_pytorch_model_promax.safetensors --relative-path models/controlnet
comfy model download --url https://huggingface.co/SG161222/RealVisXL_V5.0_Lightning/resolve/main/RealVisXL_V5.0_Lightning_fp16.safetensors --relative-path models/checkpoints

# Make sure to remove the trailing _fp16 from the RealVisXL file, as workflow.json does not expect it to be there
mv models/checkpoints/RealVisXL_V5.0_Lightning_fp16.safetensors models/checkpoints/XL_RealVisXL_V5.0_Lightning.safetensors

# Download nodes
git clone https://github.com/lquesada/ComfyUI-Inpaint-CropAndStitch.git custom_nodes/ComfyUI-Inpaint-CropAndStitch

Starting ComfyUI

When in the comfy install directory:

source venv/bin/activate
# If installed via the pypi package
comfy launch
# If installed via git
python main.py

It is recommended you remove the line not relevant to your install.

Installing the RapidRAW connector

Open a new terminal in your desired connector install directory, and install from git:

git clone https://github.com/CyberTimon/RapidRAW-AI-Connector.git
cd RapidRAW-AI-Connector/
python -m venv venv
source venv/bin/activate
pip install -r "requirements.txt"

Editing the default workflow.json

The current workflow.json file is broken, as InpaintCropImproved requires the device_mode input. To fix this you can add "device_mode": "gpu (much faster)" to its inputs.

Starting the RapidRAW connector

When in the connector install directory:

source venv/bin/activate
# Configure default port
COMFY_HOST=127.0.0.1 COMFY_PORT=8188 python main.py

Configuring RapidRAW

In RapidRAW, set the connector address to: 127.0.0.1:5000

Extras

Automation

To make it more seamless, you can edit your .desktop file to launch ComfyUI and the connector automatically. (Configure commands and directories to be correct for your install, this script assumes you installed ComfyUI from git)

Create the startup script, for example at ~/comfy/start_rapidraw_comfy.sh:

#!/bin/bash
COMFY_DIR="$HOME/comfy/ComfyUI"
CONNECTOR_DIR="$HOME/comfy/RapidRAW-AI-Connector"
COMFY_HOST=127.0.0.1
COMFY_PORT=8188

# Launch RapidRAW Flatpak
/usr/bin/flatpak run --branch=stable --arch=x86_64 --command=RapidRAW io.github.CyberTimon.RapidRAW &
RAPIDRAW_PID=$!

# Start ComfyUI
(
cd "$COMFY_DIR" || exit
source venv/bin/activate
python main.py
) &
COMFY_PID=$!

# Await ComfyUI
until (echo > /dev/tcp/$COMFY_HOST/$COMFY_PORT) >/dev/null 2>&1; do
  sleep 1
done

# Start Connector
(
cd "$CONNECTOR_DIR" || exit
source venv/bin/activate
COMFY_HOST=$COMFY_HOST COMFY_PORT=$COMFY_PORT python main.py
) &
CONNECTOR_PID=$!

wait "$RAPIDRAW_PID"
kill "$CONNECTOR_PID" 2>/dev/null
kill "$COMFY_PID" 2>/dev/null

Make it excecutable:

chmod +x ~/comfy/start_rapidraw_comfy.sh

Then edit your .desktop file so that it points to the script: (replace YOUR_USERNAME with your username)

Exec=/home/YOUR_USERNAME/comfy/start_rapidraw_comfy.sh

Note that starting the ComfyUI server can take a while (30s), so don't get alarmed if the connection test doesn't succeed immediately.

[haico1992]

Damn, we need to add this to the AI section

[lusanagi]

THIS is exactly what I was missing: the model downloading
Almost threw the towel because of that 405 error and inpainting not doing anything (though it still happens), but now I get errors because its trying to load 50GB and my 9070xt only has 16 🫠
Oh well, you win some, you lose some; I guess I'll just have to be glad I can edit on Linux now

EDIT: if I launch comfyui using the CPU it works, slow, but it works

[mmadesignerunknown]

Hi Everyone,

If possible, can anyone please guide me on how to install and use ComfyUI for generative edits on Windows 10,
I have never used ComfyUI before, so all the explanation is still kind of confusing for me.

Thanks in advance,
Regards,

[tazztone]

krita extension "ai diffusion" also uses comfyui as backend. but they use a custom node for "using ComfyUI as a backend for external tools."

krita also supports installing the comfyui managed server automatically including the needed models. maybe that approach would make it easier to use, compared to RapidRAW-AI-Connector ...