~10 min read

Generate Your First Image

~10 minBeginnerYou'll build: A 1024×1024 AI-generated image

In this tutorial you’ll install Thumper-Run, set up ComfyUI, and generate your first AI image — all in under 10 minutes (plus download time).

Requires CUDA 11.8+ or ROCm 5.7+. CPU mode is ~60x slower and not recommended for interactive use.

GPU	VRAM	Install Time	Gen Time
RTX 4090	24 GB	~3 min	~5s
RTX 3060	12 GB	~5 min	~20s
RX 7900 XT	20 GB	~5 min	~12s
AMD 780M (APU)	Shared	~8 min	~90s
CPU only	N/A	~8 min	~5 min

This tutorial assumes a GPU with 6+ GB VRAM. CPU mode works too but image generation will be significantly slower.

Step 1: Install Thumper-Run

Download the desktop app for your platform. No account or signup needed.

Linux

bash

# Download the AppImage
wget https://thumper.run/download/linux/thumper-run.AppImage
chmod +x thumper-run.AppImage
./thumper-run.AppImage

macOS

Open the .dmg from the download page, drag to Applications, then right-click → Open on first launch.

Windows

Run the .exe installer from the download page. If Windows Defender flags it, click "More info" → "Run anyway".

After launching, Thumper-Run auto-detects your GPU. You’ll see a brief "Detecting hardware..." message, then the catalog.

The ComfyUI listing on app.thumper.run: title, upstream and publisher line, category and tags — The ComfyUI listing page (captured 2026-09-14).

Step 2: Install ComfyUI

ComfyUI is a node-based image generation tool. It’s the most flexible option in the catalog.

In the catalog, find ComfyUI (or type "comfy" in the search bar)
Click the Install button on the app card
Wait for installation to complete (cloning repo, creating venv, installing dependencies)
The SDXL Starter model pack downloads automatically (~6.5 GB)

Installation typically takes 3–8 minutes depending on your internet speed and whether you already have models cached. You can monitor progress in the console: it shows each stage (cloning, creating venv, pip install, model download) with progress bars and percentages.

If installation fails at the pip step, check that you have at least 15 GB free disk space and Python 3.10+ available.

Step 3: Generate Your First Image

Time to create your first image!

Click Launch on the ComfyUI card
Wait for the health check to pass (the status changes from "Starting..." to "Running")
The default workflow loads automatically with SDXL Base
Click the blue Queue Prompt button in the bottom-right of the canvas toolbar (or press Ctrl+Enter)
Watch the progress bar — generation takes 10–60 seconds depending on your GPU

First generation loads the model into VRAM (~5–15s depending on your GPU). Subsequent runs reuse the cached model and are significantly faster.

Step 4: Customize Your Prompt

Now let’s make it your own.

Change the Prompt

Click the text in the "CLIP Text Encode (Positive)" node and replace it with your own description:

a majestic mountain landscape at sunset, golden light,
dramatic clouds, photorealistic, 8k, detailed

Adjust Settings

Steps — more steps = more detail but slower (20–30 is typical)
CFG Scale — how closely to follow the prompt (7–12 is typical)
Seed — fixed seed = reproducible results; random = variety
Resolution — 1024×1024 for SDXL (don’t go below 768 or above 1536)

Try the negative prompt node too: add things you don’t want, like "blurry, low quality, watermark".

Step 5: View in Gallery

Every image you generate is automatically saved and indexed in the Gallery.

Click the Gallery icon in the left sidebar (grid icon below the dashboard). The gallery opens showing a grid of all generated images, newest first.
Your generated images appear sorted by date (newest first)
Click any image to see full metadata: prompt, model, steps, seed, CFG, and generation time
Press S to star your favorites

Troubleshooting

Problem	Cause	Fix	Details
OOM error	Not enough VRAM	Close other GPU apps, reduce resolution to 768×768	SDXL needs ~6 GB VRAM at 1024×1024
Permission denied	AppImage not executable	Run chmod +x on the AppImage file	Linux only; macOS/Windows handle this automatically
pip timeout	Slow/unstable network	Retry install; pip caches partial downloads	The install pipeline resumes from cached packages
CUDA not found	Missing NVIDIA drivers	Install CUDA 11.8+ from nvidia.com or your package manager	Run nvidia-smi to verify driver installation

Key Takeaways

ComfyUI installs from the catalog with one click — no Python setup needed
The SDXL Starter model pack (~6.5 GB) downloads automatically on first install
First generation is slower (model loading). Subsequent runs reuse the cached model.
All generated images are saved to the Gallery with full metadata (prompt, seed, model)
Explore settings: Steps (quality), CFG (prompt adherence), Seed (reproducibility)

Next Steps

Try a chatbot — First Chatbot tutorial
Try different models — browse the model management docs
Explore other apps — try Fooocus for a simpler interface, or LTX Video for video generation
Train a LoRA on your style — Train Your First LoRA tutorial
ComfyUI deep dive — ComfyUI on Thumper guide
Read the full documentation for advanced features

Learn More

First Chatbot Tutorial

Chat with a local LLM

ComfyUI on Thumper

Advanced ComfyUI guide

User Guide

Full app management documentation

Troubleshooting

GPU issues and benchmarks