Skip to content

armgit / hypit

Public Imported from GitHub
0

Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.

fix: a Track naming five takes still fits in a Windows path

rponeb7cbd1870d3d
diff --git a/packages/hyperframes/src/document.ts b/packages/hyperframes/src/document.ts
index 68234d87..ed8adf0f 100644
--- a/packages/hyperframes/src/document.ts
+++ b/packages/hyperframes/src/document.ts
@@ -1,3 +1,5 @@
+import { createHash } from "node:crypto";
+
import { assertCompositableSurfaceRef } from "@hypit/media";
import type { CompositableSurfaceRef, FontArtifactRef } from "@hypit/media";
import { programSpaceFrameCount } from "@hypit/program-space";
@@ -197,8 +199,23 @@ function attributes(values: readonly VisualAttribute[] | undefined): string {
return (values ?? []).map(({ name, value }) => ` ${name}="${escapeHtml(value)}"`).join("");
}
+/**
+ * Name a DOM node after what it is, in a fixed number of characters.
+ *
+ * HyperFrames reads the `<video>` element's id back out of the document and spends it as a
+ * directory name for that video's extracted frames. On Windows those frames live under
+ * `%TEMP%\hf-render-…`, which leaves around 150 characters before the path stops being one. An
+ * identity that carried its parts verbatim did not fit: a Track spanning five takes names all
+ * five, the Present and the layer each repeat the Track's name, and encoding the three of them
+ * multiplied the result again. A single take was already within five characters of the limit.
+ *
+ * A digest is the same identity at a length that does not depend on how much was said. It is
+ * still stable, so the document still renders the same bytes for the same composition and the
+ * frame cache still hits. Nothing reads the parts back: `data-hypit-element-id` and
+ * `data-hypit-track-id` carry them, spelled the way an author wrote them.
+ */
function stableDomId(parts: readonly string[]): string {
- return `hypit-${Buffer.from(JSON.stringify(parts)).toString("base64url")}`;
+ return `hypit-${createHash("sha256").update(JSON.stringify(parts)).digest("hex").slice(0, 24)}`;
}
function exactFontFamily(font: FontArtifactRef): string {
diff --git a/packages/hyperframes/test/document.test.ts b/packages/hyperframes/test/document.test.ts
index a2e17e68..aecdaa3d 100644
--- a/packages/hyperframes/test/document.test.ts
+++ b/packages/hyperframes/test/document.test.ts
@@ -342,3 +342,71 @@ test("exact timed sampling lowers loop boundaries and held frames without zero-r
assert.equal((document.html.match(/data-playback-rate="1"/gu) ?? []).length, 4);
assert.doesNotMatch(document.html, /data-playback-rate="0"/u);
});
+
+test("a Track naming five takes still lowers to DOM identities a Windows path can hold", () => {
+ // HyperFrames spends the video element's id as a directory name for that video's extracted
+ // frames, under `%TEMP%\hf-render-XXXXXX\compiled\__hyperframes_video_frames\<id>\frame_%05d.png`.
+ // That surround is 103 characters of a 260-character budget, so the id is what decides whether
+ // extraction can write at all. These identities are the ones a five-take speech Track produces:
+ // the Track names every take, and the Present and the layer each repeat the Track.
+ const trackId = "speech-visual:opening-monologue-take-1+opening-monologue-take-2"
+ + "+opening-monologue-take-3+opening-monologue-take-4+opening-monologue-take-5";
+ const presentId = `${trackId}:clip-1`;
+ const programSpace = sealProgramSpace({
+ durationSec: 8 / 30,
+ frameRate: { numerator: 30, denominator: 1 },
+ });
+ const track = sealVisualTrack({
+ visualIr: "hypit.visual-ir@1",
+ id: trackId,
+ presents: [{
+ id: presentId,
+ span: { startFrame: 0, endFrameExclusive: 8 },
+ stacking: { order: 1, tieBreak: presentId },
+ elements: [{
+ id: `${presentId}:foreground`,
+ order: 0,
+ kind: "video",
+ artifact: {
+ kind: "blob" as const,
+ digest: fixtureDigest("hyperframes:long-identity-video"),
+ size: 1_000,
+ mediaType: "video/mp4",
+ },
+ muted: true,
+ // Sampling is what splits the element into parts, and each part appends its own suffix.
+ sampling: {
+ sourceFrameRate: { numerator: 30, denominator: 1 },
+ sourceFrameCount: 8,
+ segments: [
+ {
+ target: { startFrame: 0, endFrameExclusive: 4 },
+ sourceFrame: { numerator: 0, denominator: 1 },
+ rate: { numerator: 1, denominator: 1 },
+ },
+ {
+ target: { startFrame: 4, endFrameExclusive: 8 },
+ sourceFrame: { numerator: 4, denominator: 1 },
+ rate: { numerator: 1, denominator: 1 },
+ },
+ ],
+ },
+ style: [{ name: "position", value: "absolute" }, { name: "inset", value: 0 }],
+ }],
+ }],
+ });
+ const document = compileHyperframesDocument(sealComposition({
+ id: "long-identity",
+ canvas: { width: 100, height: 100, clearColor: "#000000" },
+ tracks: [track],
+ }), programSpace);
+
+ // The `id` attribute only; `data-hypit-element-id` and friends are meant to stay readable.
+ const identities = [...document.html.matchAll(/\sid="([^"]*)"/gu)].map((match) => match[1]!);
+ assert.ok(identities.length > 0, "the document names something");
+ assert.deepEqual(identities.filter((id) => id.length > 64), [],
+ "a DOM identity does not grow with the number of takes its Track names");
+ // The parts stay readable where a reader looks for them, spelled the way the author wrote them.
+ assert.ok(document.html.includes(`data-hypit-track-id="${trackId}"`));
+ assert.ok(document.html.includes(`data-hypit-element-id="${presentId}:foreground"`));
+});