I've been quietly maintaining a Fooocus2026 fork since it went LTS. It now has tag autocomplete, a job queue, an asset browser and a few other things

When Fooocus went into maintenance mode I couldn't let it go, it's still the best "type prompt, get good image" UI out there. So I forked 2.5.5 and kept adding the stuff I kept missing from a1111. A year later the list got long enough that it feels worth sharing. It's called Fooocus2026.

The two newest additions, and honestly my favorites:

• Tag autocomplete in the prompt boxes. Danbooru + e621 tag lists (the same CSVs the a1111 tagcomplete extension uses), sorted by post count, aliases included, colored by category. But it also indexes your own library: LoRA trigger words pulled from civitai metadata, your embeddings (inserted with proper syntax), your wildcards when you type __. Underscores convert to spaces on insert because SDXL checkpoints mostly want natural language, config switch available if you're on Pony and want raw tags.
• Job queue. The worker inside Fooocus was always a sequential queue, the UI just never let you use it. Now a + Queue button snapshots the full current state (prompt, model, loras, resolved seed, everything) and stacks it. Change whatever, stack again, hit run, walk away. Stop pauses the queue instead of erasing it.

Older stuff that accumulated over the months:

• Asset browser: standalone gallery for outputs plus previews for your checkpoints/loras/embeddings, with civitai preview fetching and an optional NSFW blur
• CivitAI integration: trigger words for loras AND checkpoints, community consensus settings, save any of it as a preset
• Multi-upscaler support (ESRGAN family, SwinIR, DAT, your a1111 model folders work) plus CodeFormer/GFPGAN face restoration
• Custom resolution: any aspect ratio and size, snapped to /64, no config editing
• Wildcards editor in the UI, embeddings panel, negative prompt moved next to the positive one where it belongs
• An experimental Omost integration that builds structured prompts through a local Ollama model, nothing leaves your machine

Everything I add is off by default and toggled with one line in config.txt, so a fresh clone behaves like vanilla Fooocus until you opt in. Still on gradio 3.41 like upstream. Meant for local use, don't expose it to the internet.

Code and full changelog: github.com/mikecastrodemaria/Fooocus2026

Tested on one machine (mine), so consider yourselves beta testers. If something breaks, comment or open an issue, I actually fix things. Next up is an x/y/z comparison grid built on top of the queue.