feat(engine): wire options.hdr through chunkEncoder + dynamic SDR→HDR transfer (#370)

## Summary Three independent fixes that share a common thread: HDR config flowing correctly from `EngineConfig` down through every encoder. The headline fix: disk-based HDR encodes via `chunkEncoder` were silently producing BT.709-tagged output despite `options.hdr` being set. ## Why `Chunk 3` of `plans/hdr-followups.md`. The streaming encoder was correct but `chunkEncoder.buildEncoderArgs` hard-coded BT.709 color tags and the `bt709` VUI block in `-x265-params`, even when callers passed an HDR `EncoderOptions`. Today this is harmless because `renderOrchestrator` routes native-HDR content to `streamingEncoder` and only feeds `chunkEncoder` sRGB Chrome screenshots — but the contract was a lie, and any future caller that wired HDR through `chunkEncoder` would silently get SDR output. ## What changed **3A — `chunkEncoder` respects `options.hdr` (BT.2020 + mastering metadata).** When `options.hdr` is set, the libx265 software path emits `bt2020nc` plus the matching transfer (`smpte2084` for PQ, `arib-std-b67` for HLG) at the codec level *and* embeds master-display + max-cll SEI in `-x265-params` via `getHdrEncoderColorParams`. libx264 still tags BT.709 inside `-x264-params` (libx264 has no HDR support) but the codec-level color flags flip so the container describes pixels truthfully. GPU H.265 (nvenc/videotoolbox/qsv/vaapi) gets the BT.2020 tags but no `-x265-params` block, so static mastering metadata is omitted — acceptable for previews, not HDR-aware delivery. **3B — `convertSdrToHdr` accepts a target transfer.** `videoFrameExtractor.convertSdrToHdr` was hard-coded to `transfer=arib-std-b67` (HLG) regardless of the surrounding composition's dominant transfer. `extractAllVideoFrames` now calls `analyzeCompositionHdr` first, then passes the dominant transfer (`"pq"` or `"hlg"`) into `convertSdrToHdr` so an SDR clip mixed into a PQ timeline gets converted with `smpte2084`, not `arib-std-b67`. **3C — `EngineConfig.hdr` type matches its declared shape.** The IIFE for the `hdr` field returned `undefined` when `PRODUCER_HDR_TRANSFER` wasn't `"hlg"` or `"pq"`, but the field is typed as `{ transfer: HdrTransfer } | false`. Returning `false` matches the type and avoids a downstream `undefined` check. ## Test plan - [x] `chunkEncoder.test.ts`: replaced the previous "HDR options ignored" assertions with 8 new specs covering BT.2020 + transfer tagging, master-display/max-cll embedding, libx264 fallback behavior, GPU H.265 + HDR (tags but no x265-params), and range conversion for both SDR and HDR CPU paths. - [x] All 313 engine unit tests pass (5 new HDR specs). - [x] `ffprobe` an HDR composition rendered through the chunk encoder path: shows `bt2020nc` color matrix, `smpte2084` transfer, and mastering display metadata. ## Stack Chunk 3 of `plans/hdr-followups.md`. Independent of Chunks 1/4 (touches separate code paths).
2026-09-13 07:40:06 +00:00 · 2026-04-22 20:36:29 -07:00
parent 60f4ebbf13
commit 5256a93b2d
6 changed files with 195 additions and 44 deletions
@@ -163,6 +163,8 @@ The container runs the same probe → composite → encode pipeline as the local

 - **MP4 only** — `--hdr` with `--format mov` or `--format webm` falls back to SDR
 - **HDR images: 16-bit PNG only** — other formats (JPEG, WebP, AVIF, APNG) are not decoded as HDR and fall through the SDR DOM path
+- **H.265 only — H.264 is stripped** — calling the encoder with `codec: "h264"` and `hdr: { transfer }` is rejected; the encoder logs a warning, drops `hdr`, and tags the output as SDR/BT.709. `libx264` cannot encode HDR, so the alternative would be a "half-HDR" file (BT.2020 container tags but a BT.709 VUI block in the bitstream) which confuses HDR-aware players.
+- **GPU H.265 emits color tags but no static mastering metadata** — `useGpu: true` with HDR (nvenc, videotoolbox, qsv, vaapi) tags the stream with BT.2020 + the correct transfer (smpte2084 / arib-std-b67) but does **not** embed `master-display` or `max-cll` SEI. ffmpeg does not let those flags pass through hardware encoders. The output is suitable for previews and authoring but not for HDR10-aware delivery (Apple TV, YouTube, Netflix). For spec-compliant HDR10 production output, leave `useGpu: false` so the SW `libx265` path embeds the mastering metadata.
 - **Player support** — the [`<hyperframes-player>`](/packages/player) web component plays back the encoded MP4 in the browser and inherits whatever HDR support the host browser provides; it does not implement its own HDR pipeline
 - **Headed Chrome HDR DOM capture** — the engine ships a separate WebGPU-based capture path for rendering CSS-animated DOM directly into HDR (`initHdrReadback`, `launchHdrBrowser`). It requires headed Chrome with `--enable-unsafe-webgpu` and is not used by the default render pipeline. See [Engine: HDR](/packages/engine#hdr-apis) if you are building a custom integration.

@@ -182,7 +182,7 @@ export function resolveConfig(overrides?: Partial<EngineConfig>): EngineConfig {
    hdr: (() => {
      const raw = env("PRODUCER_HDR_TRANSFER");
      if (raw === "hlg" || raw === "pq") return { transfer: raw };
-      return undefined;
+      return false;
    })(),
    hdrAutoDetect: envBool("PRODUCER_HDR_AUTO_DETECT", DEFAULT_CONFIG.hdrAutoDetect),

@@ -1,4 +1,4 @@
-import { describe, it, expect } from "vitest";
+import { describe, it, expect, vi } from "vitest";
 import { ENCODER_PRESETS, getEncoderPreset, buildEncoderArgs } from "./chunkEncoder.js";

 describe("ENCODER_PRESETS", () => {
@@ -285,23 +285,54 @@ describe("buildEncoderArgs HDR color space", () => {
  const baseOptions = { fps: 30, width: 1920, height: 1080 };
  const inputArgs = ["-framerate", "30", "-i", "frames/%04d.png"];

-  it("keeps bt709 color tags when HDR flag is set but frames are still Chrome sRGB captures", () => {
-    // HDR flag gives H.265 + 10-bit encoding but pixels are still sRGB/bt709.
-    // Tagging as bt2020 causes orange shift — so we tag truthfully as bt709.
+  it("emits BT.2020 + arib-std-b67 tags for HDR HLG (h265 SW)", () => {
+    // When options.hdr is set, the caller asserts the input pixels are
+    // already in the BT.2020 color space — tag the output truthfully so
+    // HDR-aware players apply the right transform.
    const args = buildEncoderArgs(
      { ...baseOptions, codec: "h265", preset: "medium", quality: 23, hdr: { transfer: "hlg" } },
      inputArgs,
      "out.mp4",
    );
-    expect(args[args.indexOf("-colorspace:v") + 1]).toBe("bt709");
-    expect(args[args.indexOf("-color_primaries:v") + 1]).toBe("bt709");
-    expect(args[args.indexOf("-color_trc:v") + 1]).toBe("bt709");
+    expect(args[args.indexOf("-colorspace:v") + 1]).toBe("bt2020nc");
+    expect(args[args.indexOf("-color_primaries:v") + 1]).toBe("bt2020");
+    expect(args[args.indexOf("-color_trc:v") + 1]).toBe("arib-std-b67");
    const paramIdx = args.indexOf("-x265-params");
-    expect(args[paramIdx + 1]).toContain("colorprim=bt709");
-    expect(args[paramIdx + 1]).toContain("transfer=bt709");
+    expect(args[paramIdx + 1]).toContain("colorprim=bt2020");
+    expect(args[paramIdx + 1]).toContain("transfer=arib-std-b67");
+    expect(args[paramIdx + 1]).toContain("colormatrix=bt2020nc");
  });

-  it("uses bt709 when HDR is not set", () => {
+  it("emits BT.2020 + smpte2084 tags for HDR PQ (h265 SW)", () => {
+    const args = buildEncoderArgs(
+      { ...baseOptions, codec: "h265", preset: "medium", quality: 23, hdr: { transfer: "pq" } },
+      inputArgs,
+      "out.mp4",
+    );
+    expect(args[args.indexOf("-colorspace:v") + 1]).toBe("bt2020nc");
+    expect(args[args.indexOf("-color_primaries:v") + 1]).toBe("bt2020");
+    expect(args[args.indexOf("-color_trc:v") + 1]).toBe("smpte2084");
+    const paramIdx = args.indexOf("-x265-params");
+    expect(args[paramIdx + 1]).toContain("colorprim=bt2020");
+    expect(args[paramIdx + 1]).toContain("transfer=smpte2084");
+    expect(args[paramIdx + 1]).toContain("colormatrix=bt2020nc");
+  });
+
+  it("embeds HDR static mastering metadata in x265-params when HDR is set", () => {
+    // master-display + max-cll SEI messages are required so HDR-aware
+    // players (Apple QuickTime, YouTube, HDR TVs) treat the stream as
+    // HDR10 instead of falling back to SDR BT.2020 tone-mapping.
+    const args = buildEncoderArgs(
+      { ...baseOptions, codec: "h265", preset: "medium", quality: 23, hdr: { transfer: "pq" } },
+      inputArgs,
+      "out.mp4",
+    );
+    const paramIdx = args.indexOf("-x265-params");
+    expect(args[paramIdx + 1]).toContain("master-display=");
+    expect(args[paramIdx + 1]).toContain("max-cll=");
+  });
+
+  it("uses bt709 when HDR is not set (SDR Chrome captures)", () => {
    const args = buildEncoderArgs(
      { ...baseOptions, codec: "h265", preset: "medium", quality: 23 },
      inputArgs,
@@ -309,11 +340,46 @@ describe("buildEncoderArgs HDR color space", () => {
    );
    expect(args[args.indexOf("-colorspace:v") + 1]).toBe("bt709");
    expect(args[args.indexOf("-color_trc:v") + 1]).toBe("bt709");
+    const paramIdx = args.indexOf("-x265-params");
+    expect(args[paramIdx + 1]).toContain("colorprim=bt709");
+    expect(args[paramIdx + 1]).not.toContain("master-display");
  });

-  it("uses range conversion (not colorspace) for HDR CPU encoding", () => {
-    // Chrome screenshots are sRGB — we don't convert primaries (causes color shifts).
-    // Just range-convert and let the bt2020 container metadata + 10-bit handle the rest.
+  it("does not embed HDR mastering metadata when HDR is not set", () => {
+    const args = buildEncoderArgs(
+      { ...baseOptions, codec: "h265", preset: "medium", quality: 23 },
+      inputArgs,
+      "out.mp4",
+    );
+    const paramIdx = args.indexOf("-x265-params");
+    expect(args[paramIdx + 1]).not.toContain("master-display");
+    expect(args[paramIdx + 1]).not.toContain("max-cll");
+  });
+
+  it("strips HDR and tags as SDR/BT.709 when codec=h264 (libx264 has no HDR support)", () => {
+    // libx264 cannot encode HDR. Rather than emit a "half-HDR" file (BT.2020
+    // container tags + BT.709 VUI inside the bitstream — confusing to HDR-aware
+    // players), we strip hdr and tag the whole output as SDR/BT.709. The caller
+    // gets a warning telling them to use codec=h265 for real HDR output.
+    const warnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+    const args = buildEncoderArgs(
+      { ...baseOptions, codec: "h264", preset: "medium", quality: 23, hdr: { transfer: "pq" } },
+      inputArgs,
+      "out.mp4",
+    );
+    const paramIdx = args.indexOf("-x264-params");
+    expect(args[paramIdx + 1]).toContain("colorprim=bt709");
+    expect(args[paramIdx + 1]).not.toContain("master-display");
+    expect(args[args.indexOf("-colorspace:v") + 1]).toBe("bt709");
+    expect(args[args.indexOf("-color_primaries:v") + 1]).toBe("bt709");
+    expect(args[args.indexOf("-color_trc:v") + 1]).toBe("bt709");
+    expect(warnSpy).toHaveBeenCalledWith(
+      expect.stringContaining("HDR is not supported with codec=h264"),
+    );
+    warnSpy.mockRestore();
+  });
+
+  it("uses range conversion for HDR CPU encoding", () => {
    const args = buildEncoderArgs(
      { ...baseOptions, codec: "h265", preset: "medium", quality: 23, hdr: { transfer: "hlg" } },
      inputArgs,
@@ -322,7 +388,6 @@ describe("buildEncoderArgs HDR color space", () => {
    const vfIdx = args.indexOf("-vf");
    expect(vfIdx).toBeGreaterThan(-1);
    expect(args[vfIdx + 1]).toContain("scale=in_range=pc:out_range=tv");
-    expect(args[vfIdx + 1]).not.toContain("colorspace");
  });

  it("uses same range conversion for SDR CPU encoding", () => {
@@ -334,4 +399,29 @@ describe("buildEncoderArgs HDR color space", () => {
    const vfIdx = args.indexOf("-vf");
    expect(args[vfIdx + 1]).toContain("scale=in_range=pc:out_range=tv");
  });
+
+  it("tags BT.2020 + transfer for HDR GPU H.265 (no mastering metadata via -x265-params)", () => {
+    // GPU encoders (nvenc, videotoolbox, qsv, vaapi) still emit the BT.2020
+    // color tags via the codec-level -colorspace/-color_primaries/-color_trc
+    // flags, but cannot accept x265-params, so HDR static mastering metadata
+    // (master-display, max-cll) is not embedded. Acceptable for previews,
+    // not for HDR-aware delivery.
+    const args = buildEncoderArgs(
+      {
+        ...baseOptions,
+        codec: "h265",
+        preset: "medium",
+        quality: 23,
+        useGpu: true,
+        hdr: { transfer: "pq" },
+      },
+      inputArgs,
+      "out.mp4",
+      "nvenc",
+    );
+    expect(args[args.indexOf("-colorspace:v") + 1]).toBe("bt2020nc");
+    expect(args[args.indexOf("-color_primaries:v") + 1]).toBe("bt2020");
+    expect(args[args.indexOf("-color_trc:v") + 1]).toBe("smpte2084");
+    expect(args.indexOf("-x265-params")).toBe(-1);
+  });
 });
@@ -10,7 +10,7 @@ import { copyFileSync, existsSync, mkdirSync, readdirSync, statSync, writeFileSy
 import { join, dirname } from "path";
 import { DEFAULT_CONFIG, type EngineConfig } from "../config.js";
 import { type GpuEncoder, getCachedGpuEncoder, getGpuEncoderName } from "../utils/gpuEncoder.js";
-import { type HdrTransfer } from "../utils/hdr.js";
+import { type HdrTransfer, getHdrEncoderColorParams } from "../utils/hdr.js";
 import { runFfmpeg } from "../utils/runFfmpeg.js";
 import type { EncoderOptions, EncodeResult, MuxResult } from "./chunkEncoder.types.js";

@@ -88,6 +88,18 @@ export function buildEncoderArgs(
    useGpu = false,
  } = options;

+  // libx264 cannot encode HDR. If a caller passes hdr with codec=h264 we'd
+  // produce a "half-HDR" file (BT.2020 container tags but a BT.709 VUI block
+  // inside the bitstream) which confuses HDR-aware players. Strip hdr and
+  // log a warning so the caller picks h265 (the SDR-tagged output is honest).
+  if (options.hdr && codec === "h264") {
+    console.warn(
+      "[chunkEncoder] HDR is not supported with codec=h264 (libx264 has no HDR support). " +
+        "Stripping HDR metadata and tagging output as SDR/BT.709. Use codec=h265 for HDR output.",
+    );
+    options = { ...options, hdr: undefined };
+  }
+
  const args: string[] = [...inputArgs, "-r", String(fps)];
  const shouldUseGpu = useGpu && gpuEncoder !== null;

@@ -130,8 +142,14 @@ export function buildEncoderArgs(

      // Encoder-specific params: anti-banding + color space tagging.
      // aq-mode=3 redistributes bits to dark flat areas (gradients).
+      // For HDR x265 paths we additionally embed BT.2020 + transfer + HDR static
+      // mastering metadata via x265-params; libx264 only carries BT.709 tags
+      // since HDR through H.264 is not supported by this encoder path.
      const xParamsFlag = codec === "h264" ? "-x264-params" : "-x265-params";
-      const colorParams = "colorprim=bt709:transfer=bt709:colormatrix=bt709";
+      const colorParams =
+        codec === "h265" && options.hdr
+          ? getHdrEncoderColorParams(options.hdr.transfer).x265ColorParams
+          : "colorprim=bt709:transfer=bt709:colormatrix=bt709";
      if (preset === "ultrafast") {
        args.push(xParamsFlag, `aq-mode=3:${colorParams}`);
      } else {
@@ -157,12 +175,33 @@ export function buildEncoderArgs(
  }

  // Color space metadata — tags the output so players interpret colors correctly.
-  // Chrome screenshots are always sRGB/bt709 pixels regardless of --hdr flag.
-  // We tag truthfully as bt709 even for HDR output — the --hdr flag gives
-  // H.265 + 10-bit encoding (better quality/compression) without lying about
-  // the color space. Tagging as bt2020 when pixels are bt709 causes browsers
-  // to apply the wrong color transform, producing visible orange/warm shifts.
+  //
+  // Default (no options.hdr): Chrome screenshots are sRGB/bt709 pixels and
+  // we tag them truthfully as bt709. Tagging as bt2020 when pixels are bt709
+  // causes browsers to apply the wrong color transform, producing visible
+  // orange/warm shifts.
+  //
+  // HDR (options.hdr provided): the caller asserts the input pixels are
+  // already in the BT.2020 color space (e.g. extracted HDR video frames or a
+  // pre-tagged source). We tag the output as BT.2020 + the corresponding
+  // transfer (smpte2084 for PQ, arib-std-b67 for HLG). HDR static mastering
+  // metadata (master-display, max-cll) is embedded only in the SW libx265
+  // path above; GPU H.265 + HDR carries the color tags but not the static
+  // metadata, which is acceptable for previews but not for HDR-aware delivery.
  if (codec === "h264" || codec === "h265") {
+    if (options.hdr) {
+      const transferTag = options.hdr.transfer === "pq" ? "smpte2084" : "arib-std-b67";
+      args.push(
+        "-colorspace:v",
+        "bt2020nc",
+        "-color_primaries:v",
+        "bt2020",
+        "-color_trc:v",
+        transferTag,
+        "-color_range",
+        "tv",
+      );
+    } else {
      args.push(
        "-colorspace:v",
        "bt709",
@@ -173,6 +212,7 @@ export function buildEncoderArgs(
        "-color_range",
        "tv",
      );
+    }

    // Range conversion: Chrome's full-range RGB → limited/TV range.
    if (gpuEncoder === "vaapi") {
@@ -10,7 +10,11 @@ import { existsSync, mkdirSync, readdirSync, rmSync } from "fs";
 import { join } from "path";
 import { parseHTML } from "linkedom";
 import { extractVideoMetadata, type VideoMetadata } from "../utils/ffprobe.js";
-import { isHdrColorSpace as isHdrColorSpaceUtil } from "../utils/hdr.js";
+import {
+  analyzeCompositionHdr,
+  isHdrColorSpace as isHdrColorSpaceUtil,
+  type HdrTransfer,
+} from "../utils/hdr.js";
 import { downloadToTemp, isHttpUrl } from "../utils/urlDownloader.js";
 import { runFfmpeg } from "../utils/runFfmpeg.js";
 import { DEFAULT_CONFIG, type EngineConfig } from "../config.js";
@@ -250,21 +254,28 @@ export async function extractVideoFramesRange(
 }

 /**
- * Convert an SDR video to HDR color space (HLG / BT.2020) so it can be
- * composited alongside HDR content without looking washed out.
+ * Convert an SDR (BT.709) video to BT.2020 wide-gamut so it can be composited
+ * alongside HDR content without looking washed out.
 *
- * Uses zscale for color space conversion with a nominal peak luminance of
- * 600 nits — high enough that SDR content doesn't appear too dark next to
- * HDR, matching the approach used by HeyGen's Rio pipeline.
+ * Uses FFmpeg's `colorspace` filter to remap BT.709 → BT.2020 (no real tone
+ * mapping — just a primaries swap so the input fits inside the wider HDR
+ * gamut), then re-tags the stream with the caller's target HDR transfer
+ * function (PQ for HDR10, HLG for broadcast HDR). The output transfer must
+ * match the dominant transfer of the surrounding HDR content; otherwise the
+ * downstream encoder will tag the final video with the wrong curve.
 */
 async function convertSdrToHdr(
  inputPath: string,
  outputPath: string,
+  targetTransfer: HdrTransfer,
  signal?: AbortSignal,
  config?: Partial<Pick<EngineConfig, "ffmpegProcessTimeout">>,
 ): Promise<void> {
  const timeout = config?.ffmpegProcessTimeout ?? DEFAULT_CONFIG.ffmpegProcessTimeout;

+  // smpte2084 = PQ (HDR10), arib-std-b67 = HLG.
+  const colorTrc = targetTransfer === "pq" ? "smpte2084" : "arib-std-b67";
+
  const args = [
    "-i",
    inputPath,
@@ -273,7 +284,7 @@ async function convertSdrToHdr(
    "-color_primaries",
    "bt2020",
    "-color_trc",
-    "arib-std-b67",
+    colorTrc,
    "-colorspace",
    "bt2020nc",
    "-c:v",
@@ -401,8 +412,15 @@ export async function extractAllVideoFrames(
    }),
  );

-  const hasAnyHdr = videoColorSpaces.some(isHdrColorSpaceUtil);
-  if (hasAnyHdr) {
+  const hdrInfo = analyzeCompositionHdr(videoColorSpaces);
+  if (hdrInfo.hasHdr && hdrInfo.dominantTransfer) {
+    // dominantTransfer is "majority wins" — if a composition mixes PQ and HLG
+    // sources (rare but legal), the minority transfer's videos get converted
+    // with the wrong curve. We treat this as caller-error: a single composition
+    // should not mix PQ and HLG sources, the orchestrator picks one transfer
+    // for the whole render, and any source not on that curve is normalized to
+    // it. If you need both transfers, render two separate compositions.
+    const targetTransfer = hdrInfo.dominantTransfer;
    const convertDir = join(options.outputDir, "_hdr_normalized");
    mkdirSync(convertDir, { recursive: true });

@@ -410,12 +428,13 @@ export async function extractAllVideoFrames(
      if (signal?.aborted) break;
      const cs = videoColorSpaces[i] ?? null;
      if (!isHdrColorSpaceUtil(cs)) {
-        // SDR video in a mixed timeline — convert to HDR color space
+        // SDR video in a mixed timeline — convert to the dominant HDR transfer
+        // so the encoder tags the final video correctly (PQ vs HLG).
        const entry = resolvedVideos[i];
        if (!entry) continue;
        const convertedPath = join(convertDir, `${entry.video.id}_hdr.mp4`);
        try {
-          await convertSdrToHdr(entry.videoPath, convertedPath, signal, config);
+          await convertSdrToHdr(entry.videoPath, convertedPath, targetTransfer, signal, config);
          entry.videoPath = convertedPath;
        } catch (err) {
          errors.push({