Feat/render chain v1

.cargo/config.toml

@@ -0,0 +1,10 @@
+[alias]
+build-dev = "build --profile dev-fast"
+run-dev = "run --profile dev-fast"
+
+[build]
+incremental = true
+
+[target.x86_64-unknown-linux-gnu]
+linker = "clang"
+rustflags = ["-C", "link-arg=-fuse-ld=lld"]

.claude/skills/mold/SKILL.md

@@ -178,7 +178,9 @@ mold run ltx-2-19b-distilled:fp8 "lantern-lit cave entrance" --camera-control do
 
 **Models:** `ltx-2-19b-dev:fp8`, `ltx-2-19b-distilled:fp8`, `ltx-2.3-22b-dev:fp8`, `ltx-2.3-22b-distilled:fp8`
 
-**Important flags:** `--audio`, `--no-audio`, `--audio-file`, `--video`, repeatable `--keyframe`, repeatable `--lora`, `--pipeline`, `--retake`, `--camera-control`, `--spatial-upscale`, `--temporal-upscale`
+**Important flags:** `--audio`, `--no-audio`, `--audio-file`, `--video`, repeatable `--keyframe`, repeatable `--lora`, `--pipeline`, `--retake`, `--camera-control`, `--spatial-upscale`, `--temporal-upscale`, `--clip-frames`, `--motion-tail`
+
+**Chained (arbitrary-length) video output:** for LTX-2 19B and 22B distilled models, `--frames` above the 97-frame per-clip cap automatically renders multiple clips with a motion-tail of latents carried across each clip boundary, then stitches them into a single MP4. The CLI picks this path transparently — `mold run ltx-2-19b-distilled:fp8 "a cat walking" --frames 400` produces one 400-frame MP4 from 5 chained stages. Advanced callers can override the per-clip length via `--clip-frames N` (must be `8k+1`, clamped to the model cap) and the overlap via `--motion-tail N` (default 4 pixel frames, 0 disables carryover). Chains fail closed on mid-stage failure (no partial output) and run on a single GPU. Other model families reject `--frames > 97` with an actionable error.
 
 **Current constraints:** `x2` spatial upscaling is wired across the family, `x1.5` spatial upscaling is wired for `ltx-2.3-*`, and `x2` temporal upscaling is wired in the native runtime. Camera-control preset aliases currently auto-resolve the published LTX-2 19B LoRAs only. The family runs through the native Rust stack in `mold-inference`, with CUDA as the supported backend for real local generation, CPU as a correctness-only fallback, and Metal unsupported. On 24 GB Ada GPUs such as the RTX 4090, the validated path stays on the compatible `fp8-cast` mode rather than Hopper-only `fp8-scaled-mm`. The native CUDA matrix is validated across 19B/22B text+audio-video, image-to-video, audio-to-video, keyframe, retake, public IC-LoRA, spatial upscale (`x1.5` / `x2` where published), and temporal upscale (`x2`). When requests go through `mold serve`, the built-in body limit is `64 MiB`, which is enough for common inline source-video and source-audio workflows.
 
@@ -535,6 +537,20 @@ MOLD_HOST=http://gpu-host:7680 mold run "a cat"
 MOLD_OUTPUT_DIR=/srv/mold/output mold serve
 ```
 
+### HTTP API Endpoints
+
+Core endpoints exposed by `mold serve` (full list + schemas at `/api/docs`):
+
+- `POST /api/generate` — image/video generation, raw bytes response
+- `POST /api/generate/stream` — SSE progress + base64 complete event
+- `POST /api/generate/chain` — chained arbitrary-length video (LTX-2 distilled); body is `mold_core::chain::ChainRequest` (canonical `stages[]` or auto-expand `prompt`+`total_frames`+`clip_frames`)
+- `POST /api/generate/chain/stream` — same as above, SSE progress with per-stage `denoise_step` events
+- `POST /api/expand` — LLM prompt expansion
+- `GET /api/models` · `POST /api/models/load` · `POST /api/models/pull` · `DELETE /api/models/unload`
+- `GET /api/gallery` · `GET /api/gallery/image/:name` · `GET /api/gallery/thumbnail/:name` · `DELETE /api/gallery/image/:name`
+- `POST /api/upscale` · `POST /api/upscale/stream`
+- `GET /api/status` · `GET /health` · `GET /api/capabilities`
+
 ### Prometheus Metrics
 
 When built with the `metrics` feature flag (included in Docker images and Nix builds), the server exposes a `GET /metrics` endpoint in Prometheus text exposition format. This endpoint is excluded from auth and rate limiting for monitoring scrapers.

.github/workflows/ci.yml

@@ -8,6 +8,7 @@ on:
 
 env:
   CARGO_TERM_COLOR: always
+  CARGO_INCREMENTAL: "1"
 
 # Note: CUDA builds are not run in CI — they require a GPU host with NixOS + CUDA.
 # CI only checks non-CUDA compilation, lints, and tests.
@@ -56,60 +57,76 @@ jobs:
         working-directory: web
         run: bun run build
 
-  fmt:
+  rust:
     runs-on: ubuntu-latest
+    env:
+      RUSTC_WRAPPER: sccache
+      SCCACHE_GHA_ENABLED: "true"
     steps:
       - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@stable
         with:
-          components: rustfmt
-      - run: cargo fmt --all -- --check
-
-  check:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v6
-      - uses: dtolnay/rust-toolchain@stable
+          components: rustfmt,clippy
+      - name: Install Rust build deps
+        run: sudo apt-get update && sudo apt-get install -y clang lld nasm libwebp-dev
+      - uses: mozilla-actions/sccache-action@v0.0.7
+      - name: Probe sccache (disable on cache outage)
+        shell: bash
+        run: |
+          set +e
+          printf 'fn main() {}\n' > /tmp/sccache_probe.rs
+          sccache rustc --edition 2021 -- /tmp/sccache_probe.rs -o /tmp/sccache_probe
+          status=$?
+          rm -f /tmp/sccache_probe /tmp/sccache_probe.rs
+          if [ $status -ne 0 ]; then
+            echo "sccache probe failed — disabling RUSTC_WRAPPER for this job"
+            echo "RUSTC_WRAPPER=" >> "$GITHUB_ENV"
+            sccache --stop-server >/dev/null 2>&1 || true
+          fi
       - uses: Swatinem/rust-cache@v2
         with:
           shared-key: workspace-default
           save-if: ${{ github.ref == 'refs/heads/main' }}
+      - name: Format
+        run: cargo fmt --all -- --check
       - name: Check
         run: cargo check --workspace
-
-  clippy:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v6
-      - uses: dtolnay/rust-toolchain@stable
-        with:
-          components: clippy
-      - uses: Swatinem/rust-cache@v2
-        with:
-          shared-key: workspace-default
-          save-if: ${{ github.ref == 'refs/heads/main' }}
       - name: Clippy
-        run: cargo clippy --workspace -- -D warnings
-
-  test:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v6
-      - uses: dtolnay/rust-toolchain@stable
-      - uses: Swatinem/rust-cache@v2
-        with:
-          shared-key: workspace-default
-          save-if: ${{ github.ref == 'refs/heads/main' }}
+        run: cargo clippy --workspace --all-targets -- -D warnings
       - name: Test
         run: cargo test --workspace
+      - name: Check with all optional features
+        run: cargo check -p mold-ai --features preview,discord,expand,tui,webp,mp4
 
   coverage:
     runs-on: ubuntu-latest
+    env:
+      RUSTC_WRAPPER: sccache
+      SCCACHE_GHA_ENABLED: "true"
     steps:
       - uses: actions/checkout@v6
 
       - uses: dtolnay/rust-toolchain@stable
 
+      - name: Install Rust build deps
+        run: sudo apt-get update && sudo apt-get install -y clang lld nasm libwebp-dev
+
+      - uses: mozilla-actions/sccache-action@v0.0.7
+
+      - name: Probe sccache (disable on cache outage)
+        shell: bash
+        run: |
+          set +e
+          printf 'fn main() {}\n' > /tmp/sccache_probe.rs
+          sccache rustc --edition 2021 -- /tmp/sccache_probe.rs -o /tmp/sccache_probe
+          status=$?
+          rm -f /tmp/sccache_probe /tmp/sccache_probe.rs
+          if [ $status -ne 0 ]; then
+            echo "sccache probe failed — disabling RUSTC_WRAPPER for this job"
+            echo "RUSTC_WRAPPER=" >> "$GITHUB_ENV"
+            sccache --stop-server >/dev/null 2>&1 || true
+          fi
+
       - uses: Swatinem/rust-cache@v2
         with:
           shared-key: workspace-llvm-cov
@@ -127,17 +144,3 @@ jobs:
           files: lcov.info
           token: ${{ secrets.CODECOV_TOKEN }}
           fail_ci_if_error: false
-
-  check-features:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v6
-      - uses: dtolnay/rust-toolchain@stable
-      - uses: Swatinem/rust-cache@v2
-        with:
-          shared-key: workspace-all-features
-          save-if: ${{ github.ref == 'refs/heads/main' }}
-      - name: Install system deps for optional features
-        run: sudo apt-get update && sudo apt-get install -y nasm libwebp-dev
-      - name: Check with all features (preview, discord, expand, tui, webp, mp4)
-        run: cargo check -p mold-ai --features preview,discord,expand,tui,webp,mp4

.github/workflows/release.yml

@@ -12,11 +12,17 @@ concurrency:
 jobs:
   build-macos:
     runs-on: macos-14
+    env:
+      CARGO_INCREMENTAL: "1"
+      RUSTC_WRAPPER: sccache
+      SCCACHE_GHA_ENABLED: "true"
     steps:
       - uses: actions/checkout@v6
 
       - uses: dtolnay/rust-toolchain@stable
 
+      - uses: mozilla-actions/sccache-action@v0.0.7
+
       - uses: Swatinem/rust-cache@v2
         with:
           key: macos-release
@@ -39,11 +45,15 @@ jobs:
   # Ada Lovelace (RTX 40-series, sm_89)
   build-linux-sm89:
     runs-on: ubuntu-latest
+    env:
+      CARGO_INCREMENTAL: "1"
+      RUSTC_WRAPPER: sccache
+      SCCACHE_GHA_ENABLED: "true"
     steps:
       - uses: actions/checkout@v6
 
-      - name: Install nasm (required by openh264 source build for mp4 feature)
-        run: sudo apt-get update && sudo apt-get install -y nasm
+      - name: Install build deps
+        run: sudo apt-get update && sudo apt-get install -y clang lld nasm
 
       - uses: Jimver/cuda-toolkit@v0.2.35
         with:
@@ -53,6 +63,8 @@ jobs:
 
       - uses: dtolnay/rust-toolchain@stable
 
+      - uses: mozilla-actions/sccache-action@v0.0.7
+
       - uses: Swatinem/rust-cache@v2
         with:
           key: linux-cuda-sm89-release
@@ -75,11 +87,15 @@ jobs:
   # Blackwell (RTX 50-series, sm_120)
   build-linux-sm120:
     runs-on: ubuntu-latest
+    env:
+      CARGO_INCREMENTAL: "1"
+      RUSTC_WRAPPER: sccache
+      SCCACHE_GHA_ENABLED: "true"
     steps:
       - uses: actions/checkout@v6
 
-      - name: Install nasm (required by openh264 source build for mp4 feature)
-        run: sudo apt-get update && sudo apt-get install -y nasm
+      - name: Install build deps
+        run: sudo apt-get update && sudo apt-get install -y clang lld nasm
 
       - uses: Jimver/cuda-toolkit@v0.2.35
         with:
@@ -89,6 +105,8 @@ jobs:
 
       - uses: dtolnay/rust-toolchain@stable
 
+      - uses: mozilla-actions/sccache-action@v0.0.7
+
       - uses: Swatinem/rust-cache@v2
         with:
           key: linux-cuda-sm120-release

.gitignore

@@ -6,6 +6,7 @@
 # Claude Code
 .claude/worktrees/
 .claude/scheduled_tasks.lock
+.worktrees/
 .playwright-mcp/
 .direnv/