AIpubAIpub

AI 101 · beginner field guide

Getting started with AI

A calm first route into local AI image-making: choose one tool, make one image, and learn the vocabulary without pretending the rabbit hole is shallow.

Start from zero Ask AIpub community

No sign-in, wallet, or mystery installer required to read this guide.

AIpub's green neon ramen-shop scene, featuring the words Getting started with AI

Step one

Choose your tool, not your entire future

You do not need every interface. Pick the one that helps you make a first image with the least friction, then learn its basics before collecting new launchers like cursed Pokémon cards.

Recommended for power & flexibility

ComfyUI

A visual node-and-workflow interface for people who want to see every stage of an image pipeline.

  • Reusable workflows
  • Deep control
  • Grows with your projects

Know before you install: The learning curve is real. Start with a template workflow and change one node at a time.

Official guide

Easy traditional WebUI

Forge

A prompt-first WebUI route for people who prefer familiar controls over a node graph.

  • Prompt-first screen
  • Simple first render
  • Many classic WebUI tutorials apply

Know before you install: The packaged upstream route is specifically CUDA/PyTorch-oriented; use its current docs for your hardware.

Official guide

The classic

AUTOMATIC1111

The longstanding Stable Diffusion WebUI with a huge archive of older tutorials and extensions.

  • Familiar tutorial language
  • Prompt-first workflow
  • Lots of historical references

Know before you install: Use current project instructions, not a random old video—setup and compatibility details change.

Official guide

Easiest package manager

Stability Matrix

A launcher and package manager that can install and organize several compatible image tools.

  • Guided installs
  • Separate package environments
  • Shared model library

Know before you install: It is not a model or generator itself. Choose an actual UI package inside it, such as ComfyUI or a WebUI.

Official guide

Step two

Install one route cleanly

These are signposts, not copied installer scripts. Follow the official project page for your operating system and hardware; it is more current than a screenshot or a random video description.

ComfyUI: Desktop first, portable when it fits

Recommended path: for a supported Windows or macOS setup, start with ComfyUI Desktop. It handles the initial Python environment and helps you arrive at a starter workflow without a manual clone.

  1. Read the project’s system requirements for compatibility caveats.
  2. Use the Desktop guide on Windows, or follow the official documentation for your platform.
  3. Start with a default/template workflow, choose a compatible model, then use the first-generation guide .
  4. For Desktop, use Help → Open Folder → Open Model Folder instead of guessing an absolute model path.

The Windows Portable package is an official alternative. Manual installs and custom nodes are advanced paths—save them until your starter workflow works.

Forge: a traditional WebUI route

Forge’s upstream README documents a packaged one-click route. It is a traditional WebUI choice, especially for people who prefer prompt controls over a graph.

  1. Download the appropriate package from Forge’s upstream releases .
  2. Extract it to a folder you can find again.
  3. Run update.bat, then run run.bat, as the upstream README directs.
  4. Open the local address the launcher shows and make a first image before adding extensions.

This packaged route is not universal hardware advice. If your GPU or operating system does not match its current guidance, use the upstream Forge documentation or choose a managed package route.

AUTOMATIC1111: the classic tutorial route

AUTOMATIC1111 remains useful when a tutorial specifically names it. Its manual Windows route expects Python 3.10.6 and Git, then launches through webui-user.bat.

  1. Start at the official repository , not a repackaged download.
  2. For a current NVIDIA-specific path, read the project’s installation guide .
  3. Keep the initial setup plain. Only add an extension after the base UI starts and creates a normal image.

Hardware support varies. Treat older tutorials as historical context and verify launch flags, Python versions, and backend guidance with the project itself.

Stability Matrix: manage packages, not one giant mystery folder

Stability Matrix is a manager and launcher. It can install separate UI packages, choose an appropriate backend, and connect them to a shared model library.

  1. Download it from the official Stability Matrix downloads page .
  2. On Windows, extract the official zip and run StabilityMatrix.exe.
  3. Choose a data directory. Portable Mode keeps the library near the app and is documented as a practical default.
  4. Open Packages → Add Package, choose one UI, and accept the project’s appropriate default backend unless you know why you need another.

Read the official Add Package guide and data-directory guide before relocating or sharing a library.

Step three

The small vocabulary that saves a lot of confusion

You do not need to memorize this all today. These words help you read a model card, follow a workflow, and ask a more useful question.

Checkpoint / base model
The large model that does the main image-making work. Its model family matters when choosing add-ons and workflows.
LoRA
A smaller adapter that nudges a compatible base model toward a subject, style, or concept. A useful analogy: the checkpoint is the game; the LoRA is a mod, not a whole replacement game.
VAE
A helper that translates between the model's internal image space and pixels. Some checkpoints include one; others expect a separate compatible VAE.
Prompt & negative prompt
The prompt describes what you want. A negative prompt describes things you want to avoid when the chosen model and UI support it; it is guidance, not a guarantee.
Seed
A starting random number. Keeping the seed and settings lets you compare changes more fairly and revisit a direction later.
Sampler & CFG
The sampler is the denoising method. CFG is prompt-guidance strength. Defaults are a fine place to begin; change one setting at a time.
Workflow
A saved recipe of models, settings, and connections. In ComfyUI it is a node graph; in a classic WebUI it may be a settings recipe rather than a graph.

Step four

Models are files, choices, and responsibilities

A model host is a catalogue, not a compatibility guarantee. Slow down long enough to check the author’s description, base model, license, and file type.

  • Prefer a known project or creator, and read the model card before downloading.
  • Confirm that a LoRA or workflow matches your checkpoint/base-model family.
  • Keep source, version, license, and trigger notes with the file so future-you is not haunted.
  • Do not run unknown executables, scripts, or “fixers” bundled with a model download.
  • Use .safetensors where the author provides it, but still verify the source and compatibility.

Start with model pages on Civitai or Hugging Face ; both are hosts, not endorsements of every upload.

Step five

Know where the files live

These are common default-relative locations, not promises. Managed installs can point elsewhere; use the application’s own folder/library controls before creating new directories by hand.

Typical model locations
Tool or installCheckpointLoRAVAEImportant note
ComfyUI portable/manualComfyUI/models/checkpointsComfyUI/models/lorasComfyUI/models/vaeUse extra_model_paths.yaml only when you intentionally share a library.
ComfyUI DesktopUse Help → Open Folder → Open Model Folder.Desktop manages its own locations; do not assume a portable path.
Traditional WebUI (A1111 / Forge)models/Stable-diffusionmodels/Loramodels/VAEThese defaults can be overridden by launch configuration.
Stability MatrixModels/StableDiffusionModels/LoraModels/VAEThese live under your selected library; let package setup link them.

If a model does not appear, refresh or restart the UI, then check its expected folder and base-model compatibility. The ComfyUI model guide is a strong reference for model types and shared paths.

Step six

Make a first image, then try one compatible LoRA

Your first-image recipe

  1. Choose one compatible checkpoint and use the UI’s starter workflow or defaults.
  2. Write a short prompt with a subject, setting, and visual direction: “a tiny observatory in rain, ink-and-watercolour illustration, soft window light”.
  3. Generate once, save the image, and record the prompt, model, seed, and workflow/settings.
  4. Change only one thing for the next image: a phrase, seed, sampler, resolution, or guidance value.

Then add a LoRA

  1. Read the LoRA author’s compatibility and trigger-word notes.
  2. Put it in the correct folder or import it through your chosen manager.
  3. Use the interface’s own LoRA picker/loader so it applies the correct syntax or node.
  4. Start modestly, compare against the baseline, and adjust one strength at a time.

When it gets weird

Troubleshoot without escalating the chaos

Keep the last error, your UI version, operating system, GPU/backend, model name, and the latest change. That tiny receipt turns “it broke” into something another human can help solve.

The app will not start

Return to the project’s official guide, check the exact first error, and verify that the download matched your platform. Do not replace Python, Torch, or launch files from a random comment until you understand why.

It runs out of memory or freezes

Use the starter workflow, lower resolution or batch size, and avoid adding multiple large models at once. Larger models and higher-resolution workflows require more memory; compatibility is tool- and backend-specific.

The model or LoRA does not appear

Check the file location, restart or refresh the UI, and confirm the model family matches the loader/workflow. For managed installs, use the manager’s model-library controls.

The output looks broken or ignores my idea

Go back to a simple compatible checkpoint and starter workflow. Remove extra LoRAs, keep the seed, then reintroduce one change at a time.

An extension or custom node fails

Disable or remove the last add-on only if the project’s own instructions describe how. Keep the original error and ask the extension maintainer or AIpub before applying broad dependency changes.

A gentler learning path

Seven small steps beat one enormous setup

  1. Choose one interface that matches how you like to learn.

  2. Install only that interface, using its own current guide.

  3. Make one image with a supplied or starter workflow before adding extras.

  4. Record the model, seed, prompt, and settings that made something interesting.

  5. Change one variable at a time: prompt, seed, sampler, resolution, or a compatible LoRA.

  6. Save a tiny workflow or settings recipe you can repeat tomorrow.

  7. Share a work-in-progress and ask one focused question in AIpub.

Keep this page current

Official resources and what each one is for

Installation advice ages quickly. The links below are centralized in this site and were last reviewed on . Follow the project’s current documentation over an old tutorial.

You are allowed to ask beginner questions

Bring the first weird result, not a perfect portfolio.

Share a screenshot, your tool, model, prompt, and the exact thing that surprised you. That is plenty to start a useful conversation.

Join AIpub Discord Browse the forum