From 07e3a04df9e44840008b28c1f2915029ee8dcf11 Mon Sep 17 00:00:00 2001 From: Koorosh Aslansefat Date: Thu, 24 Sep 2026 07:37:25 -0700 Subject: [PATCH 1/2] docs: add illustrated image editing SMILE workflow --- .../blogs/image-editing-smile-concept.svg | 78 +++++++++++++++++++ 1 file changed, 78 insertions(+) create mode 100644 docs/assets/images/blogs/image-editing-smile-concept.svg diff --git a/docs/assets/images/blogs/image-editing-smile-concept.svg b/docs/assets/images/blogs/image-editing-smile-concept.svg new file mode 100644 index 0000000..7b07085 --- /dev/null +++ b/docs/assets/images/blogs/image-editing-smile-concept.svg @@ -0,0 +1,78 @@ + + An image-editing instruction and its local word explanation + A city street and the instruction Make the weather snowy enter an image-editing model. The edited street is snowy. Below, SMILE compares edits after changing instruction words to estimate local word contributions. The emphasis is illustrative, not a measured result. + + + + + + + + + + + + + + + A + Input image + instruction + + + + + + + + + + + + + + Prompt: Make the weather snowy + + + + + + Image-editing + model + B + + + + C + Edited image + + + + + + + + + + + + + + + + + + Same street, snowy weather + + + + D + SMILE estimates which instruction words shaped the edit + Change selected words, compare the resulting images, then fit a local explanation. + + Make + the + weather + snowy + + Illustrative emphasis only · actual contributions depend on the model, image and comparison + From 8279fe5d097d86415b43ac535d3a7d1eaea847fc Mon Sep 17 00:00:00 2001 From: Koorosh Aslansefat Date: Thu, 24 Sep 2026 07:37:32 -0700 Subject: [PATCH 2/2] docs: explain image editing SMILE with diagram --- docs/research/reddit/blogs/image-editing-smile.md | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/docs/research/reddit/blogs/image-editing-smile.md b/docs/research/reddit/blogs/image-editing-smile.md index e4e0d47..57a59ee 100644 --- a/docs/research/reddit/blogs/image-editing-smile.md +++ b/docs/research/reddit/blogs/image-editing-smile.md @@ -13,6 +13,12 @@ description: How changing a few words in an image-editing instruction can help e An image-editing model can follow an instruction such as “replace the cloudy sky with a clear blue sky,” but a convincing result does not tell us how strongly each part of the instruction shaped the edit. That question matters when a visual change could hide something important, particularly in applications involving medical or safety-related images. +The diagram below traces one example: a street image and a request for snowy weather go into an image editor, which produces a winter version. SMILE then tests changes to the instruction and estimates how much individual words matter to this particular edit. + +![A street image and a request for snowy weather enter an image-editing model; the snowy result is compared with edits from changed instructions to estimate local word contributions. The highlighted words are illustrative, not measured values.](../../../assets/images/blogs/image-editing-smile-concept.svg) + +*Schematic example. Word emphasis illustrates a possible explanation, not measured results.* + ## Change the instruction and watch the output The image-editing SMILE approach makes controlled changes to the text instruction. It removes or changes selected words, runs the editor again, and compares each result with the reference edit. A local model then estimates which parts of the instruction are associated with the largest changes. The explanation can be shown as contributions for individual terms rather than a single score for the whole prompt.