vast.ai

ToBs

is it also possible to use vast.ai instead of runpod.io?
Normally I use your tutorial and runpod. But in the last days, runpod is extremly slow.

I think the comfy-ui-cloud is too expensive and only availible by pay by month and not "by use"?

KairosJS

ToBs I've never tried Vast.ai, but I don't see why it wouldn't work.

I've been using Novita.ai as a replacement for RunPod lately, I've also been facing a lot of issues with their services, unfortunately.
And the 4090 GPU is very cheap: ₀.35€

ToBs

I dont really know how to Set it up.
With your tutorial for runpod it was easy.
Maybe you can Do one for novita?

KairosJS

ToBs I made a quick video to show you, but overall it's exactly the same as runpod.io:

ToBs

Thank you!
But I was looking for a tutorial to use it for Image-Creation.
Or maybe is it possible to run image and text-generation on the same instance (runpod or vast.ai)?

KairosJS

ToBs Quick video.

As before you want to instantiate a GPU pod (pick a 4090 or 5090).

This time use pytorch 2.8.0 as template.
Open the 8188 port (for ComfyUI)
Depending on the number of models you want to download, increase the volume to 100GB
Deploy
Once it's running, open the 8888 port to use jupyter and open a terminal

In this terminal use the following commands:

apt update
apt install nano
nano deploy.sh

Then copy the following script that will install and download all you need to run comfyui with Anima or Krea2.
Feel free to change the models used below to your likings. I've added:
DaSiWa Krea2V2TurboBF16.safetensors that is a Krea2 Turbo finetune that I like
One obsession_AnimaV3.safetensors that is an Anima finetune that I like

qwen3vl_4b_fp8_scaled.safetensors is the clip model for Krea2
qwen_3_06b_base.safetensors is the clip model for Anima

Adapt the script below with you civitai token and the model you want to use.

# KREA2 + ANIMA:
#!/bin/bash
git clone https://github.com/Comfy-Org/ComfyUI.git
cd ComfyUI
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu130
pip install -r requirements.txt
cd models/diffusion_models/
wget -O "DaSiWa Krea2V2TurboBF16.safetensors" --show-progress "https://civitai.red/api/download/models/3151280?fileId=3034252&token=[civitai_token]"
wget -O "One obsession_AnimaV3.safetensors" --show-progress "https://civitai.red/api/download/models/3190113?fileId=3070857&token=[civitai_token]"
cd ../clip/
wget https://huggingface.co/Comfy-Org/Krea-2/resolve/main/text_encoders/qwen3vl_4b_fp8_scaled.safetensors
wget https://huggingface.co/circlestone-labs/Anima/resolve/main/split_files/text_encoders/qwen_3_06b_base.safetensors
cd ../vae/
wget https://huggingface.co/Comfy-Org/Krea-2/resolve/main/vae/qwen_image_vae.safetensors
cd ../../
python main.py --listen --enable-cors-header="*"

Once it's pasted, close it with ctrl + x, then press "y"

then:

sh deploy.sh

It will take quite some time to download everything. Once it's working, copy the comfyui URL into eudaimonia.