Image generators, upscalers and the open source stacks you can run yourself. Each entry says where the model is strong, because they are much less interchangeable than the marketing suggests.
Still the best default aesthetic in the category, which is why art directors keep coming back. It gives you less literal control than the alternatives, so it is a poor fit when a client has an exact layout in mind.
Reachable inside ChatGPT, so you can describe an edit in plain language and iterate conversationally. Prompt understanding is excellent; fine detail and texture are not its best qualities.
Home of the Stable Diffusion model family, whose open weights built most of the ecosystem below. Choose it when you need to run models yourself or fine-tune on your own images.
Black Forest Labs models that beat most rivals on prompt following, anatomy and legible text in an image. It has quietly become the default base model for serious open source workflows.
Aimed at game and product art, with trained model presets and consistent character tools. The generous free credits make it a cheap way to learn image prompting.
The one to use when the image needs readable words in it: posters, logos, packaging mockups. Photographic realism is weaker than the leaders.
Real-time canvas where the image updates as you sketch or type. The fastest way to explore a direction before committing, and it also does solid upscaling.
An upscaler that invents plausible detail rather than just adding pixels, which is remarkable on portraits and product shots. Expensive, and it will hallucinate detail that was never there.
Bundles several generation models with a huge stock library and licensing you can point a legal team at. Convenient rather than best in class on any single model.
Trained on licensed content and covered by Adobe indemnification, which is the whole reason brands pick it. Output quality trails Midjourney and Flux.
Generates real vector output and holds a consistent style across a set, so icon sets and illustration systems come out usable. Designers pick it over Midjourney when the asset has to be edited afterwards.
Browser canvas built around graphics rather than fine art: social posts, thumbnails, simple layouts. Good free allowance for casual use.
Node-based workflow editor for running diffusion models locally with total control. The learning curve is steep and there is nothing else that lets you build repeatable pipelines like this.
The classic local interface for Stable Diffusion, with an extension for everything. Less actively developed than it was, but still the gentlest entry point to running models on your own GPU.
The library where open model checkpoints and LoRAs live, with example prompts attached to every image. Essential if you run models locally; moderate the content filters before sharing your screen.
Runs Stable Diffusion and friends entirely on an iPhone, iPad or Mac with no account and no cloud. Slower than a desktop GPU, and completely private.