Image generation

Making pictures from a prompt, and the software that drives it.

Almost everything interesting in this category runs on your own machine. That is the whole reason it is the biggest section here: a picture generated locally never reaches anyone else's server, and the tool cannot change its mind about what you are allowed to make.

What you need is a graphics card with enough memory. Use the filter to see what will actually run on yours.

13 tools

ComfyUI

A node graph for image and video models. Steeper to learn than anything else here, and the only one that lets you build exactly the pipeline you want.
- Self-hosted
- Allowed
Front-ends and workflows 4 GB VRAM, 12 GB recommended
Stable Diffusion WebUI

The original web interface for Stable Diffusion, by AUTOMATIC1111. Enormously extended, widely copied, and no longer moving quickly.
- Self-hosted
- Allowed
Front-ends and workflows 4 GB VRAM, 8 GB recommended
AnimateDiff

Motion added to an existing image model. The technique that made short animated clips possible on a home graphics card.
- Self-hosted
- Allowed
Video 8 GB VRAM, 12 GB recommended
Civitai

The image model hub: checkpoints, LoRAs and the example images that show what they actually do. Indispensable, and volatile.
- Hosted
- Opt-in
Finding models Not known
Draw Things

Image generation as a native Mac and iPhone app. Free, local, and startlingly good at running large models on Apple silicon.
- Self-hosted
- Allowed
- No GPU needed
Front-ends and workflows No GPU needed
Fooocus

Image generation with almost every setting hidden. Type a prompt, get a good picture. No longer actively developed.
- Self-hosted
- Abandoned
- Allowed
Front-ends and workflows 4 GB VRAM, 8 GB recommended
InvokeAI

A polished, professional-feeling image studio with a unified canvas. The one to show someone who bounced off the others.
- Hybrid
- Allowed
Front-ends and workflows 6 GB VRAM, 12 GB recommended
Kohya's GUI

The standard tool for training your own LoRA. Unglamorous, widely used, and the thing every tutorial eventually points at.
- Self-hosted
- Allowed
Image generation 8 GB VRAM, 16 GB recommended
Krita AI Diffusion

Generation built into a real painting program. You paint, it fills in; you paint over it, it adjusts.
- Self-hosted
- Allowed
Front-ends and workflows 4 GB VRAM, 8 GB recommended
NovelAI

A subscription writing and image service with an explicit policy of not filtering fiction, and a long record of keeping it.
- Hosted
- Allowed
Writing Not known
OneTrainer

A newer trainer with a friendlier interface than Kohya's and a good record of supporting new architectures quickly.
- Self-hosted
- Allowed
Image generation 8 GB VRAM, 16 GB recommended
SD.Next

Another descendant of the AUTOMATIC1111 interface, aimed at breadth: many model families, and AMD and Intel GPUs as first-class targets.
- Self-hosted
- Allowed
Front-ends and workflows 4 GB VRAM, 8 GB recommended
Stable Diffusion WebUI Forge

A fork of AUTOMATIC1111's interface, reworked underneath for speed and lower memory use. The same screen, faster, on less VRAM.
- Self-hosted
- Allowed
Front-ends and workflows 4 GB VRAM, 8 GB recommended