Edit photos by describing the change, on your own GPU. A free, open-source alternative to Nano Banana and ChatGPT image editing: photos stay on your computer, no daily limit. One installer for Windows, Mac and Linux sets up the best open model (Qwen-Image 2.1). MCP server for AI agents.
TypeScript ai-photo-editorcomfyui-alternativegenerative-fillimage-editingimage-generationinpaintinglinuxlocal-aimacosmcpmcp-servernano-banana-alternativeofflinephoto-editorprivacyqwen-imageqwen-image-2-1stable-diffusion-cppvulkanwindows
Commits on master
What's new 0.11.1: the Mac engine, the Q4_K_M image model step, the UI fixes
ae8e479 Gabriele Lozupone committed 25 days ago
Q4_K_M as the default image model; earlier installs keep the Q4_K until they accept
2af83dd Gabriele Lozupone committed 25 days ago
mac_bench test: compare the report path in the platform's form (failed on Windows)
dcf6ce1 Gabriele Lozupone committed 25 days ago
Merge branch 'macos-optimization': Kokon's own Mac engine
7c3e256 Gabriele Lozupone committed 25 days ago
Revert: Try it turns the Mac engine on and stays where it is (Gabriele preferred it)
ee0f943 Gabriele Lozupone committed 25 days ago
Mac engine offer: Try it opens Settings → Performance (the switch on, the Benchmark button)
deda712 Gabriele Lozupone committed 25 days ago
Mac engine inside Kokon.app, offered once at start on Macs
3be7141 Gabriele Lozupone committed 25 days ago
STATUS: RDNA2 flash-attention workaround, next update's fixes
b45c4ca Gabriele Lozupone committed 25 days ago
UI fixes from GitHub and Reddit reports for the next release
076efab Gabriele Lozupone committed 25 days ago
Mac engine: prefetch after each image measured, not kept
49f9f20 Gabriele Lozupone committed 25 days ago
Mac engine: read the models back into the file cache after each image (below 32 GB)
ce9185f Gabriele Lozupone committed 25 days ago
Mac engine bench: where each 1024² image's time goes
35e03af Gabriele Lozupone committed 25 days ago
Benchmark this Mac: the Mac engine's bench from Settings → Performance
87d5919 Gabriele Lozupone committed 25 days ago
STATUS: Mac engine step, decoder and vision tower faster
2ebf5f4 Gabriele Lozupone committed 25 days ago
Mac engine: VAE mid stage as operations, compiled operations shared by every instance
9460753 Gabriele Lozupone committed 25 days ago
Mac engine: row kernels pack the Neural Engine's rows, residual fused with the LN after it
840fa90 Gabriele Lozupone committed 25 days ago
Mac engine README: the Neural Engine lanes during the attention, measured and not kept
116670a Gabriele Lozupone committed 25 days ago
Mac engine: the vision tower's heads on a 72-wide attention kernel, 1 MP encode 2.6 → 1.7 s
71fb5eb Gabriele Lozupone committed 25 days ago
Mac engine: tools/ab.sh, alternated A/B of build snapshots
77f0d49 Gabriele Lozupone committed 25 days ago
STATUS: Mac engine does brushed edits and extending, 1 MP edit ~106 s
59b41bb Gabriele Lozupone committed 25 days ago
Mac engine: brushed edits and extending an image (init image, mask, strength)
1320a5a Gabriele Lozupone committed 25 days ago
Mac engine: 1 MP edits on 16 GB, int8 prefix cache and a smaller working memory
16c5034 Gabriele Lozupone committed 25 days ago
Mac engine: edits with reference images, and Prompt+ looking at them
f60f53e Gabriele Lozupone committed 25 days ago
Mac engine: the text encoder with images, checked against transformers
415ee8f Gabriele Lozupone committed 25 days ago
Mac engine: the vision tower, checked against transformers
76d0604 Gabriele Lozupone committed 25 days ago
Mac engine: the VAE encoder, checked against diffusers
dfd6c50 Gabriele Lozupone committed 25 days ago
Mac engine: int8 Neural Engine with finer scales measured and dropped
fb9cc5f Gabriele Lozupone committed 26 days ago
STATUS: Mac engine 64-67 s per 1 MP image, tested from the app
3086518 Gabriele Lozupone committed 26 days ago
Mac engine: VAE up blocks as single operations, the last two in tiles
1dafe36 Gabriele Lozupone committed 26 days ago