Merge pull request #68 from outis1one/claude/focused-maxwell-3kgh3x

Make model prefetching automatic in install/run scripts

README.md

@@ -11,7 +11,7 @@ git clone https://github.com/outis1one/editmaskwithai
 cd editmaskwithai
 
 # One-time setup: installs nvidia-container-toolkit, configures Docker,
-# and sets up a permanent DNS fix so the container can download models.
+# sets up a permanent DNS fix, and prefetches all AI models on the host.
 chmod +x install-local-gpu.sh
 ./install-local-gpu.sh
 
@@ -22,7 +22,7 @@ chmod +x bring-up-local-gpu.sh
 
 Open **http://localhost:3080**
 
-**First startup downloads the AI model for your GPU (~13 GB, one time).** Models are cached in `./data/hf_cache/` and survive rebuilds.
+**Models (~13 GB total, one time) download automatically on the host**, outside Docker — both scripts call `./prefetch-models.sh` for you, since in-container DNS is unreliable on some hosts. They're cached in `./data/hf_cache/` and `./data/models/`, and survive rebuilds.
 
 ---
 
@@ -163,7 +163,7 @@ EditmaskwithAI/
 ├── Dockerfile.gpu
 ├── install-local-gpu.sh         # One-time GPU host setup
 ├── bring-up-local-gpu.sh        # Start/stop the GPU container
-├── prefetch-models.sh           # Download AI models on the host (DNS-blocked workaround)
+├── prefetch-models.sh           # Download AI models on the host (called automatically; also runnable standalone)
 └── .env.example
 ```
 
@@ -232,7 +232,7 @@ You can also pick a specific model per-edit from the Remove Background dialog's
 
 **AI models not downloading (container DNS blocked)**
 
-Easiest fix: download the models on the host instead of inside the container — they land in `./data/`, which is already bind-mounted into the container, so it picks them up with no rebuild:
+`install-local-gpu.sh` and `bring-up-local-gpu.sh` already run this for you automatically on every start, so you normally don't need to think about it. If a model still didn't download (no network at the time, etc.), re-run it manually — it lands in `./data/`, which is already bind-mounted into the container, so it's picked up with no rebuild:
 
 ```bash
 ./prefetch-models.sh           # SAM + U2Net + BEN2 + BiRefNet-HR (~1.5GB)

bring-up-local-gpu.sh

@@ -4,9 +4,11 @@
 # Run this each time you want to start the app.
 # Run ./install-local-gpu.sh once first on a new machine.
 #
-# If models fail to download inside the container (DNS/firewall blocked),
-# run ./prefetch-models.sh first — it downloads them on the host into
-# ./data/, which this container already bind-mounts, so no rebuild needed.
+# Before starting, this fetches any missing models on the host (outside
+# Docker) via ./prefetch-models.sh — in-container DNS/network is unreliable
+# on some hosts, so this is the default now, not a manual troubleshooting
+# step. It never blocks startup: if it fails (no network, no python3, etc.)
+# the container still starts and falls back to its own in-container download.
 #
 # Usage:
 #   ./bring-up-local-gpu.sh              # start (detached, rebuild if needed)
@@ -29,6 +31,9 @@ cd "$(dirname "$0")"
 mkdir -p data/models data/hf_cache data/projects data/patches
 
 if [ $# -eq 0 ]; then
+    # Best-effort: fetch any missing models on the host first (see header).
+    ./prefetch-models.sh --sdxl \
+        || echo "⚠ Model prefetch had failures (see above) — continuing anyway, the container will retry in-container."
     exec docker compose -f docker-compose.gpu.yml up -d --build
 else
     exec docker compose -f docker-compose.gpu.yml "$@"

install-local-gpu.sh

@@ -170,12 +170,31 @@ else
     echo "  Minimum driver version: 525"
 fi
 
+# ── 6. Prefetch all AI models (host-side, outside Docker) ───────────────────
+# In-container DNS/network is unreliable on some hosts, so downloading on the
+# host up front (including SDXL/inpaint, ~13GB) is the default now rather
+# than a manual troubleshooting step. Best-effort: this script must finish
+# (and leave the GPU/Docker setup done) even if prefetch fails outright.
+# Run as the real invoking user, not root, so downloaded files (and any
+# `pip install --user` side effects) end up owned by that user — this script
+# itself is already running as root via the sudo re-exec above.
+echo ""
+echo "Prefetching AI models (this can take a while for SDXL, ~13GB)..."
+if [ -n "${SUDO_USER:-}" ]; then
+    sudo -u "$SUDO_USER" -H "$REPO_DIR/prefetch-models.sh" --sdxl \
+        || echo "⚠ Model prefetch had failures (see above) — continuing anyway, the container will retry in-container."
+else
+    "$REPO_DIR/prefetch-models.sh" --sdxl \
+        || echo "⚠ Model prefetch had failures (see above) — continuing anyway, the container will retry in-container."
+fi
+
 echo ""
 echo "=================================================="
 echo "  Setup complete."
 echo "  Start the app with:  ./bring-up-local-gpu.sh"
 echo ""
-echo "  If models fail to download inside the container (DNS/firewall"
-echo "  blocked), fetch them on the host first instead:"
-echo "    ./prefetch-models.sh"
+echo "  Models are prefetched automatically by this script and by"
+echo "  bring-up-local-gpu.sh on every start. If any failed above (no"
+echo "  network, etc.), re-run manually any time:"
+echo "    ./prefetch-models.sh --sdxl"
 echo "=================================================="