diff --git a/main.go b/main.go
index 591afb8..92e984a 100644
--- a/main.go
+++ b/main.go
@@ -1,18 +1,22 @@
 package main
 
 import (
+	"bufio"
+	"encoding/hex"
 	"flag"
 	"fmt"
 	"image"
 	"image/color"
 	_ "image/gif"
 	_ "image/jpeg"
-	_ "image/png"
+	"image/png"
 	"log"
 	"math/rand"
 	"net"
+	"net/textproto"
 	"os"
 	"runtime/pprof"
+	"strings"
 	"time"
 )
 
@@ -25,6 +29,7 @@ var connections = flag.Int("connections", 4, "Number of simultaneous connections
 var address = flag.String("host", "127.0.0.1:1337", "Server address")
 var runtime = flag.String("runtime", "60s", "exit after timeout")
 var shuffle = flag.Bool("shuffle", false, "pixel send ordering")
+var fetchImgPath = flag.String("fetch-image", "", "path to save the fetched pixel state to")
 
 func main() {
 	flag.Parse()
@@ -52,15 +57,34 @@ func main() {
 
 	offset := image.Pt(*image_offsetx, *image_offsety)
 	img := readImage(*image_path)
+	fetchedImg := image.NewNRGBA(img.Bounds().Add(offset))
+
+	if *fetchImgPath != "" {
+		// using img.SubImage to distribute tasks is nice, as we can also parallelize command generation easily!
+		// @cleanup: use a box tiling algo instead of hardcoding
+		b := fetchedImg.Bounds()
+		b1 := image.Rectangle{b.Min, b.Size().Div(2)}
+		b2 := b1.Add(image.Pt(b1.Dx(), 0))
+		b3 := b1.Add(image.Pt(0, b1.Dy()))
+		b4 := b1.Add(b1.Size())
+		go fetchImage(fetchedImg.SubImage(b1).(*image.NRGBA))
+		go fetchImage(fetchedImg.SubImage(b2).(*image.NRGBA))
+		go fetchImage(fetchedImg.SubImage(b3).(*image.NRGBA))
+		go fetchImage(fetchedImg.SubImage(b4).(*image.NRGBA))
+		*connections -= 4
+	}
+
 	// Generate and split messages into equal chunks
 	commands := genCommands(img, offset)
 	if *shuffle {
 		shuffleCommands(commands)
 	}
 
-	commandGroups := chunkCommands(commands, *connections)
-	for _, messages := range commandGroups {
-		go bomb(messages)
+	if *connections > 0 {
+		commandGroups := chunkCommands(commands, *connections)
+		for _, messages := range commandGroups {
+			go bomb(messages)
+		}
 	}
 
 	// Terminate after timeout to save resources
@@ -69,6 +93,10 @@ func main() {
 		log.Fatal("Invalid runtime specified: " + err.Error())
 	}
 	time.Sleep(timer)
+
+	if *fetchImgPath != "" {
+		writeImage(*fetchImgPath, fetchedImg)
+	}
 }
 
 func bomb(messages []byte) {
@@ -102,6 +130,17 @@ func readImage(path string) (img image.Image) {
 	return img
 }
 
+func writeImage(path string, img image.Image) {
+	f, err := os.Create(path)
+	if err != nil {
+		log.Fatal(err)
+	}
+	if err := png.Encode(f, img); err != nil {
+		f.Close()
+		log.Fatal(err)
+	}
+}
+
 // Creates message based on given image
 func genCommands(img image.Image, offset image.Point) (commands [][]byte) {
 	b := img.Bounds()
@@ -149,3 +188,47 @@ func shuffleCommands(slice [][]byte) {
 		slice[i], slice[j] = slice[j], slice[i]
 	}
 }
+
+func fetchImage(img *image.NRGBA) {
+	// FIXME @speed: this is unusably s l o w w w
+	// bottleneck seems to be our pixel reading/parsing code. cpuprofile!
+	// -> should buffer it just as in bomb()
+
+	conn, err := net.Dial("tcp", *address)
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer conn.Close()
+
+	reader := bufio.NewReader(conn)
+	tp := textproto.NewReader(reader)
+
+	b := img.Bounds()
+	for {
+		for x := b.Min.X; x < b.Max.X; x++ {
+			for y := b.Min.Y; y < b.Max.Y; y++ {
+				// request pixel
+				fmt.Fprintf(conn, "PX %d %d\n", x, y)
+				if err != nil {
+					log.Fatal(err)
+				}
+
+				// read pixel
+				// @speed try to run this in a separate goroutine?
+				// we probably need to buffer the responses then
+				res, err := tp.ReadLine()
+				if err != nil {
+					log.Fatal(err)
+				}
+				res2 := strings.Split(res, " ")
+				col, _ := hex.DecodeString(res2[3])
+				img.Set(x, y, color.NRGBA{
+					uint8(col[0]),
+					uint8(col[1]),
+					uint8(col[2]),
+					255,
+				})
+			}
+		}
+	}
+}