diff --git a/.github/workflows/test-docker-sd.yaml b/.github/workflows/test-docker-sd.yaml
new file mode 100644
index 00000000..0413dd5c
--- /dev/null
+++ b/.github/workflows/test-docker-sd.yaml
@@ -0,0 +1,89 @@
+name: docker-test-stablediffusion
+
+on:
+  push:
+    branches:
+      - main
+    paths-ignore:
+      - '**.md'
+  pull_request:
+    branches:
+      - main
+    paths-ignore:
+      - '**.md'
+
+permissions: read-all
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    timeout-minutes: 240
+    steps:
+      - name: Harden Runner
+        uses: step-security/harden-runner@eb238b55efaa70779f274895e782ed17c84f2895 # v2.6.1
+        with:
+          egress-policy: audit
+
+      - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+
+      - name: cleanup disk space
+        run: |
+          docker system prune -f -a --volumes
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf "/usr/local/share/boost"
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+
+      # need containerd image store for testing local images
+      - uses: crazy-max/ghaction-setup-docker@69b561f709cbd934060981d481ccfc316cfb77b7 # v2.2.0
+        with:
+          daemon-config: |
+            {
+              "debug": true,
+              "features": {
+                "containerd-snapshotter": true
+              }
+            }
+      - uses: crazy-max/ghaction-github-runtime@b3a9207c0e1ef41f4cf215303c976869d0c2c1c4 # v3.0.0
+
+      - name: build aikit
+        run: |
+          docker buildx build . -t aikit:test \
+            --load --progress plain \
+            --cache-from=type=gha,scope=aikit \
+            --cache-to=type=gha,scope=aikit,mode=max
+
+      - name: build test model
+        run: |
+          docker buildx build . -t testmodel:test \
+            -f test/aikitfile-sd.yaml \
+            --load --progress plain \
+            --cache-from=type=gha,scope=testmodel \
+            --cache-to=type=gha,scope=testmodel,mode=max
+
+      - name: list images
+        run: docker images
+
+      - name: run test model
+        run: docker run --name testmodel -d -p 8080:8080 testmodel:test
+
+      - name: install e2e dependencies
+        run: make test-e2e-dependencies
+
+      - name: run test
+        run: |
+          curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
+            "model": "llava",
+            "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://discovery.sndimg.com/content/dam/images/discovery/fullset/2021/12/1/GettyImages-1280574445.jpg.rend.hgtvcom.966.644.suffix/1638387352203.jpeg" }}], "temperature": 0.9}]}'
+
+      - name: save logs
+        if: always()
+        run: |
+          docker logs testmodel > docker.log
+
+      - name: publish logs
+        if: always()
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32 # v3.1.3
+        with:
+          name: docker-logs
+          path: docker.log
diff --git a/README.md b/README.md
index aff353c2..07e300c1 100644
--- a/README.md
+++ b/README.md
@@ -18,7 +18,7 @@ AIKit uses [LocalAI](https://localai.io/) under-the-hood to run inference. Local
 - 🚀 [Easy to use declarative configuration](./docs/specs.md)
 - ✨ OpenAI API compatible to use with any OpenAI API compatible client
 - 🖼️ [Multi-modal model support](#vision-with-llava)
-- 🚢 Kubernetes deployment ready
+- 🚢 [Kubernetes deployment ready](#kubernetes-deployment)
 - 📦 Supports multiple models with a single image
 - 🖥️ [Supports GPU-accelerated inferencing with NVIDIA GPUs](#nvidia)
 - 🔐 [Signed images for `aikit` and pre-made models](./docs/cosign.md)
diff --git a/pkg/aikit2llb/convert.go b/pkg/aikit2llb/convert.go
index 1fc6e7ff..8cf71af2 100644
--- a/pkg/aikit2llb/convert.go
+++ b/pkg/aikit2llb/convert.go
@@ -22,31 +22,39 @@ const (
 
 func Aikit2LLB(c *config.Config) (llb.State, *specs.Image) {
 	var merge llb.State
-	s := llb.Image(debianSlim)
-	s, merge = copyModels(c, s)
-	s, merge = addLocalAI(c, s, merge)
+	state := llb.Image(debianSlim)
+	base := getBaseImage(c)
+
+	state, merge = copyModels(c, base, state)
+	state, merge = addLocalAI(c, state, merge)
+
+	// install cuda if runtime is nvidia
 	if c.Runtime == utils.RuntimeNVIDIA {
-		merge = installCuda(s, merge)
+		merge = installCuda(state, merge)
 	}
+
+	// install opencv and friends if stable diffusion backend is being used
 	for b := range c.Backends {
 		if strings.Contains(c.Backends[b], "stablediffusion") {
-			merge = installOpenCV(s, merge)
+			merge = installOpenCV(state, merge)
 		}
 	}
+
 	imageCfg := NewImageConfig(c)
 	return merge, imageCfg
 }
 
-func copyModels(c *config.Config, s llb.State) (llb.State, llb.State) {
-	var base llb.State
+func getBaseImage(c *config.Config) llb.State {
 	for b := range c.Backends {
 		if strings.Contains(c.Backends[b], "stablediffusion") {
-			// due to too many dependencies, we are using debian slim as base for stable diffusion
-			base = llb.Image(debianSlim)
-		} else {
-			base = llb.Image(distrolessBase)
+			// due to too many dependencies, using debian slim as base for stable diffusion
+			return llb.Image(debianSlim)
 		}
 	}
+	return llb.Image(distrolessBase)
+}
+
+func copyModels(c *config.Config, base llb.State, s llb.State) (llb.State, llb.State) {
 	savedState := s
 
 	// create config file if defined
@@ -162,7 +170,7 @@ func addLocalAI(c *config.Config, s llb.State, merge llb.State) (llb.State, llb.
 	opts = append(opts, llb.Chmod(0o755))
 	localAI := llb.HTTP(localAIURL, opts...)
 	s = s.File(
-		llb.Copy(localAI, "local-ai", "/usr/bin"),
+		llb.Copy(localAI, "local-ai", "/usr/bin/local-ai"),
 		llb.WithCustomName("Copying "+fileNameFromURL(localAIURL)+" to /usr/bin"), //nolint: goconst
 	)