Don AI Art HOLBER: Stable Battle Setup (Part 2)

Welcome to my Home Lab trip, where we are making a powerful AI art setup that is affected by epic dunes Sta! In Part 1, we set a lightweight Ukur example on our home lab machine. Now, in Part 2, we will enable remote access, create permanent containers, and call for AI art with stable dispersion. Whether you are a former soldier of Himalab or curious early, this guide will help you use The power of the desert Amazing AI-inflatory art to make. Let’s dive!

To ensure my home lab machine, dubbed ixIt is always accessible, I assigns it to the static IP (192.168.100.42) in my router DHCP settings. This prevents IP changes after reboot, which makes remote access reliable.

Next, I made access to SSH by creating a configuration file. How’s it:

1 – created (Or edit) $HOME/.ssh/config On your local machine:

Host ix
Hostname 192.168.100.42
User core
IdentityFile $HOME/.ssh/homelab

2 – Fix the permit On $HOME/.ssh/config :

chmod 600 $HOME/.ssh/config

3 – Test SSH Changing Ix’s place in the name of your Holamab Machine:

ssh ix

Now, typing SSH IX immediately adds me – don’t need to remember IPS or keys!

4 – Enable cockpit service (Useful for remote system management):

sudo systemctl enable --now cockpit.service

Pro -tip: If you use FlatopicInstall Cockpit client With:

flatpak install flathub org.cockpit_project.CockpitClient

Open the cockpit client, enter your machine name in the host field, and you are inside!

Why does it make a difference: Saving a stable IP and SSH Configuration time and reducing frustration, let you focus on entertainment equipment – such as AI art.

We need to keep the reboot and logout alive as we need to run AI services as permanent containers such as stable, halama, and kamfai. This is how I have arranged the podman for the consumer level.

1-Restart the user-level container:

mkdir -p $HOME/.config/systemd/user
cp /lib/systemd/system/podman-restart.service $HOME/.config/systemd/user
systemctl --user enable podman-restart.service

2 – The certified service is active:

systemctl --user list-unit-files | grep podman-restart

I searched the Podman Resistor Dot Service that should be made as a mark, and that was.

3 – Actable delay for session perseverance:

loginctl enable-linger $UID

Enables 4-system level (For a root -driven services):

sudo systemctl enable podman-restart.service

What’s going on? These orders ensure that your AI container is resumed automatically, as Freeman Sites endured the Arakis storms. At the user level, perseverance is ideal for personal projects, while the system level is for important services.

What are your user -level containers used for? Are you operating any system level services in your home lab? Tell me down!

When I first logged into the cockpit, I targeted a snag: Ublue-NVCTK-Cdi.service failed, which indicates a problem with my NVIDIA EGPU setup (Swant Breaky Box 750Ex with RTX 3060). Here’s how I debugged it:

1 – Check whether the NVIDIA driver is loaded:

lsmod | grep nvidia

It returned empty, so I knew the driver was not filled.

2 – tried to manually load the driver:

sudo modprobe nvidia

It failed a “key” by service “error, pointing to the colonel driver, signing a secure boot criminal.

3 – EGPU Hardware Confirmation:

rpm-ostree install -A bolt
boltctl | grep -i status

I wanted to reject any problem with the EGPU because I had not used this hardware before, and some Thunderbolt devices need permission. The above mentioned boltctl Command developed the following output:

├─ status: authorized
├─ status: authorized

Seeing the two authorized places, the issues related to the EGPU were rejected.

4 – Fixed NVIDIA drivers by inserting Ukur Mock for a safe boot

sudo mokutil --import /etc/pki/akmods/certs/akmods-ublue.der

Note: Akmods-ullate.der The aforementioned file needs to be registered because the secure boot does not allow pre -built -in MOD (drivers) to load without proper signing keys. This is a good thing.

The aforementioned command will indicate for the password and will need a re -boot. You will need to enter the password you have been assigned to complete the key process, and then you can continue to boot. Once the system surfaced, a quick spot check showed that the driver was filled with:

(core@ix ~)$ lsmod|grep nvidia
nvidia_uvm           3997696  0
nvidia_drm            151552  0
nvidia_modeset       1830912  1 nvidia_drm
nvidia              97161216  2 nvidia_uvm,nvidia_modeset
drm_ttm_helper         16384  3 nvidia_drm,xe
video                  81920  3 xe,i915,nvidia_modeset

5 – Verified Podman has access to GPU

podman run --rm --device nvidia.com/gpu=all \
--privileged nvidia/cuda:11.0.3-base-ubuntu20.04 \
nvidia-smi

If you run the command above, you should see something like that:

Success! The driver version 570.144 RTX was also listed with 3060 device. This system was eventually ready to read the burden of AI’s work.

Why keep a safe boot? It only protects your system, ensuring the signed code runs, and reduces the risks. Think of it as a digital shield against the seekers of Herconnan Hunter.

Have you faced safe boot challenges with your home lab? Share your corrections in the comments!

Now for the main event: Running a steady to create Sta-PRIMED AI ART. I did my clone AI-HOMELAB repository And made a container picture (if you are following the series please use the Branch Part 2 For this article):

git clone git@github.com:thecodenomad/ai-homelab.git
cd ai-homelab
podman build . -t sd-ix

There were something for the eager Key Highlights It came out of the construction of the containerfile (otherwise go to the firewall setting):

Twenty image nvidia/cuda:12.8.1-base-ubuntu22.04 To support GPU

Why? Ubuntu 22.04 uses the default 3.10. This is preferred because a stable dispersion, piturich, is artistic with more than 3.10 versions. The use of Nvidia containers also helped to minimize cases of installing CUDA.

Installation of rust

RUN curl --proto '=https' --tlsv1.2 -sSf  | sh -s -- -y --default-toolchain stable
ENV PATH="/home/1000/.cargo/bin:${PATH}"

Why? The stable luh The rust compiler is essential Transformer Dependent

Setup Setup

Nothing should be surprising in this section. There is a small magic to pin a stable covenant to ensure the working environment. The most notable section of this section is colliding with transformers version 4.51.3 to Address an existing issue.

RUN sed -i 's/transformers==4.30.2/transformers==4.51.3/' requirements.txt

Firewall configuration

Note: Stable dispersion runs on the port 7860So make sure to allow this TCP Traffic for Port 7860. This can be easily done by the cockpit client in the network tab. Click on Edit rules and zones Button, and then Add Services Button to enter the details:

Downloading a model and running a container

Stable dissolution requires a pre-trained model to work on its magic-this is primarily a “guided book” in which AI has learned to transform your text indicators into photos. Think about it as a device that helps to understand stable dispersion to what visuals make your words match your words, such as a guide that connects your thoughts to this art. Before we can do anything with install, we will need to keep a model.

From the inside of ai-homelab Download folder, basic v1-5-crynd-emaonly.safetensors from the model The surface of the throat:

curl -Lo models/Stable-diffusion/v1-5-pruned-emaonly.safetensors https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors

Once it expires, start the container with the following:

podman run --privileged --rm -it \
--device nvidia.com/gpu=all \
--volume ./models:/app/models:z \
--volume ./outputs:/app/outputs:z \
-p 7860:7860 \
localhost/sd-ix:latest

Once you see the line “Model Filled” in the Output, you should be able to access the Stable Battle Weboi from your Home Lab IP (such as, And be presented with similar UI:

A Chris Kine, once pulling, should taste the blood before it is sheets. – Don, Frank Herbert

Using the Web UI, I entered to enter these gestures StaThamed masterpiece:

Immediate: Underneath a wide deserted twin sun, an endless hill that shining with spices rotating is orange, remote frameen Setch, sand, cinema lighting, hypertens, 8K resolution.

Negative indicator: Fade pin, low quality, cartooning, excessive, distorted, extra organs, unnatural colors, text samples, watermark, low resolution, pixlated, non -realistic proportion, modern buildings, unsafe elements, sheep forming

Settings: Schedule type = karas, CFG = 8.0, steps = 30, and default for everything.

Conclusion: Like the unexpected winds of the members, by forming its dunes, the stable does not always follow your indicators – it creates an artistic interpretation based on its basic model and training data. My output occupied the essence of the Arakis with a wide desert, a flareing sun, and shining orange spices, but some details (such as the frames of the Seyche’s precise place and a sand worm emerging from the sand) are different.

This variation in AI art is normal: the model’s creativity adds a layer of surprise, like spices! To get closer to your vision, you can experience with different indicators, settings or seeds.

Note: To discover this specific output, set the price of the seed 3704657981. The seeds help you understand how the settings affect the results, but nevertheless, the results of the model may be slightly different.

What StaWill you make art developed? Share your indications or results in the comments! Bonus: Which AI art tool is most enthusiastic to you – stable dispersion, comphyui, or anyone else?

• Start Seach Tao: Mastering in stable IPS and SSH Configs makes you an important part of the Seich, which saves you time for creative acquisitions like AI art.
• Steel suitcare: The establishment of Podman’s resume services makes you a master in running your AI tools easily, just as a frameman ensures that his steel suit never fails.
• Tahidi Challenger: NVIDIA resolving drivers prove your flexibility, which makes you a respected supporter in the management of AI art and beyond the Himalab system.
• Spice visionary: The use of stable dispersion turns you into creative genius, channels the spice view to bring you StaArt ready for life with your home lab.

In Part 3, we will discover using podman-compose To configure comfyui (An alternative to photo generation, node -based Ui), BesetAnd maybe even produce a sand worm. Follow me on the medium for Part 3, and let’s keep the spice flowing!

Are questions? Leave them down – we can meet them together!