AI Gallery

Start with what AI Gallery is not: it is a connector to a community-run generation network. It doesn't operate the workers, train their models, choose which worker handles your job or control the image that comes back. That single architectural fact explains most of the product's behaviour and most of its quirks. What it does well is remove friction. There is no install, no account, and the Studio opens straight into Text to Image, Image to Image, Inpainting and ControlNet. The catalog lists 100+ Stable Diffusion community models, and the list is ranked by output quality in your current mode before current speed — a change from the earlier speed-first ordering that matters more than it sounds, because the model choice drives the look of the image far more than the prompt does. Modes are handled honestly: models that can't perform Image to Image or Inpainting sink to the bottom with a note, and Generate offers a one-tap switch rather than letting you fail. Editing is where it has grown most. Pro inpainting adds zoom, lasso, box and smart color select, plus mask invert, edge softening and a mask preview so you can see exactly what you're painting before you commit. You can seed Image to Image and Inpainting from a dropped file, a pasted image or any earlier creation, with presets like Subtle or Reimagine dialling how much changes. Fixed seeds let you iterate reproducibly. Outside the render loop, the November 2026 additions are practical housekeeping: palette extraction with HEX/CSS/JSON export, KB-targeted compression that produces a fitting copy without touching the original, transparent-edge trimming for logos and stickers, and a Contact Sheet Maker that combines up to 12 images into one PNG or JPEG with captions and ordering — originals untouched and nothing uploaded. Storage is the honest weakness to understand before you invest hours. The gallery lives in IndexedDB on your device, which is genuinely private and genuinely fragile: clearing browser data deletes it permanently. Backups now store WebP exactly as stored, verify every file on restore, accept multiple ZIPs at once and keep what was saved if an import runs out of space — good engineering, but you still have to actually run the backups. Rights are model-aware with a prompt to check the model, source materials and legal terms before each use; you also pick the SFW/NSFW flag and are responsible for reviewing, deleting or reporting outputs. Where it fits: personal projects, blog illustrations, social posts, wallpapers, learning to prompt, and quick mask edits where opening a full editor is overkill. Where it doesn't: client work on a deadline where queue times must be predictable, teams needing shared workspaces or an API, anyone who needs to train or fine-tune a model, and high-volume runs where consistent output matters more than cost. The note on the site that current no-fee access isn't a promise of permanent free features is worth taking at face value when you plan around it.