Changelog

Newest first.

v0.1 (coming soon)

The first public release.

Generate

• Supports SDXL (Pony, Illustrious, NoobAI, including V-Pred), Anima, FLUX.1 (dev, schnell, Kontext), FLUX.2 klein (4B, 9B), Qwen Image 2.1, Z-Image, Krea 2 and SD 1.5.
• Split models find a matching VAE and text encoder automatically; missing parts can be downloaded from official sources in Settings.
• Tag mode for prompts, Chinese tag search (about 45,000 tags translated), dynamic syntax, wildcards and variables.
• LoRAs, embeddings, reference images (ControlNet for SDXL), recipes, scheduled generation, automatic face fix after generating.
• When graphics memory is short: VRAM strategies, storing the main model in fp8, and streaming weights automatically when the model doesn’t fit.

Explore, compare, recombine

• Find images with a similar composition or mood, including your Eagle library (read-only).
• Comparison sets: side by side, slider, diff table, tournament, XY tables.
• Take a winning image apart and mix its parts with others.

Refine and output

• img2img, inpainting, face fix, upscaling, outpainting, background removal.
• One-sentence edits (Qwen Image 2.1, FLUX.2 klein, FLUX.1 Kontext).
• Layer editor, collages, watermarks, .kra export, batch export.

Library and assets

• Scans images generated by CoCo and other image generators; search by model, LoRA or prompt.
• Asset library with covers, notes, favourites, trigger words and compatibility.

Training

• Datasets, WD14 tagging, JoyCaption descriptions.
• SDXL and Anima LoRAs and slider LoRAs, with test images during training sent to a comparison set for picking an epoch.