diff --git a/README.md b/README.md
index bfd50cf..f5edc51 100644
--- a/README.md
+++ b/README.md
@@ -5,9 +5,9 @@ rga is a tool to recursively search for text in many different types of files. I
 [![Linux build status](https://api.travis-ci.org/phiresky/ripgrep_all.svg)](https://travis-ci.org/phiresky/ripgrep_all)
 [![Crates.io](https://img.shields.io/crates/v/ripgrep_all.svg)](https://crates.io/crates/ripgrep_all)
 
-# todo
+# Future Work
 
-- jpg adapter (based on object classification / detection (yolo?)) for fun
+- photograph adapter (based on object classification / detection (yolo?)) for fun, based on something [like this](https://github.com/aimagelab/show-control-and-tell). Tried, but very hard to integrate (especially state of the art approaches).
 - 7z adapter (couldn't find a nice to use rust library)
 
 # considerations
diff --git a/src/args.rs b/src/args.rs
index c63f847..1dd23dd 100644
--- a/src/args.rs
+++ b/src/args.rs
@@ -49,6 +49,7 @@ pub struct RgaArgs {
     /// so users sometimes use any or no extension at all. With this flag, rga
     /// will try to detect the mime type of input files using the magic bytes
     /// (similar to the `file` utility), and use that to choose the adapter.
+    /// Detection is only done on the first 8KiB of the file, since we can't always seek on the input (in archives).
     pub accurate: bool,
 
     #[serde(default, skip_serializing_if = "is_default")]