Skip to content

Validate video OCR image bytes before Tesseract #882

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
saurabhhhcodes wants to merge 1 commit into from
Closed

Validate video OCR image bytes before Tesseract #882

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
74 changes: 74 additions & 0 deletions src/__tests__/api/video-generate-route.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,74 @@
import { POST } from "@/app/api/video/generate/route";

jest.mock("@clerk/nextjs/server", () => ({
currentUser: jest.fn().mockResolvedValue({
id: "user_123",
primaryEmailAddress: { emailAddress: "student@example.com" },
}),
}));

jest.mock("@/lib/ratelimit/api-rate-limit", () => ({
enforceApiRateLimit: jest.fn().mockResolvedValue(null),
}));

jest.mock("@/lib/auth/auth-utils", () => ({
checkUserBlock: jest.fn().mockResolvedValue({
isBlocked: false,
errorResponse: undefined,
dbUser: undefined,
}),
}));

jest.mock("@/configs/db", () => ({
db: {
insert: jest.fn(),
},
}));

jest.mock("@/configs/schema", () => ({
videoJobsTable: {},
}));

jest.mock("@/inngest/client", () => ({
inngest: {
send: jest.fn(),
},
}));

jest.mock("@/lib/ratelimit/ratelimit", () => ({
videoLimiter: {
limit: jest.fn(),
},
redisClient: {
set: jest.fn(),
del: jest.fn(),
},
}));

describe("video generate route", () => {
it("rejects malformed image content before queueing OCR work", async () => {
const fetchSpy = jest.spyOn(globalThis, "fetch").mockResolvedValueOnce(
new Response("not an image", {
status: 200,
headers: { "Content-Type": "image/jpeg" },
}),
);

const res = await POST(
new Request("http://localhost/api/video/generate", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
imageUrl: "https://example.com/malformed.jpg",
}),
}),
);

expect(fetchSpy).toHaveBeenCalledTimes(1);
expect(res.status).toBe(422);
await expect(res.json()).resolves.toEqual({
error: "Please upload a valid PNG, JPG, or WEBP image.",
code: "INVALID_IMAGE_PAYLOAD",
});
});
});
45 changes: 45 additions & 0 deletions src/__tests__/lib/video-image-validation.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
import {
detectVideoImageMimeType,
validateVideoImageUrl,
} from "@/lib/video/image-validation";

describe("video image validation", () => {
it("detects PNG, JPEG, and WEBP magic bytes", () => {
expect(
detectVideoImageMimeType(
new Uint8Array([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]),
),
).toBe("image/png");
expect(detectVideoImageMimeType(new Uint8Array([0xff, 0xd8, 0xff, 0x00]))).toBe(
"image/jpeg",
);
expect(
detectVideoImageMimeType(
new Uint8Array([
0x52, 0x49, 0x46, 0x46, 0x00, 0x00, 0x00, 0x00, 0x57, 0x45, 0x42, 0x50,
]),
),
).toBe("image/webp");
});

it("rejects non-image payloads before OCR", async () => {
const fetchSpy = jest.spyOn(globalThis, "fetch").mockResolvedValueOnce(
new Response("not an image", {
status: 200,
headers: { "Content-Type": "image/jpeg" },
}),
);

const result = await validateVideoImageUrl("https://example.com/malformed.jpg");

expect(fetchSpy).toHaveBeenCalledWith("https://example.com/malformed.jpg", {
cache: "no-store",
});
expect(result).toEqual({
ok: false,
status: 422,
code: "INVALID_IMAGE_PAYLOAD",
error: "Please upload a valid PNG, JPG, or WEBP image.",
});
});
});
14 changes: 14 additions & 0 deletions src/app/api/video/generate/route.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
import { enforceApiRateLimit } from "@/lib/ratelimit/api-rate-limit";
import { parseAndValidateRequest } from "@/lib/validations/validate";
import { generateVideoSchema } from "@/lib/validations/video";
import { validateVideoImageUrl } from "@/lib/video/image-validation";
import { db } from "@/configs/db";
import { videoJobsTable } from "@/configs/schema";
import { inngest } from "@/inngest/client";
Expand Down Expand Up @@ -42,6 +43,19 @@
const { errorResponse, data } = await parseAndValidateRequest(req, generateVideoSchema);
if (errorResponse) return errorResponse;

if (data.imageUrl) {
const imageValidation = await validateVideoImageUrl(data.imageUrl);
if (!imageValidation.ok) {
return NextResponse.json(
{
error: imageValidation.error,

Check failure on line 51 in src/app/api/video/generate/route.ts

View workflow job for this annotation

GitHub Actions / TypeScript Check 

Property 'error' does not exist on type 'VideoImageValidationResult'.
code: imageValidation.code,

Check failure on line 52 in src/app/api/video/generate/route.ts

View workflow job for this annotation

GitHub Actions / TypeScript Check 

Property 'code' does not exist on type 'VideoImageValidationResult'.
},
{ status: imageValidation.status },

Check failure on line 54 in src/app/api/video/generate/route.ts

View workflow job for this annotation

GitHub Actions / TypeScript Check 

Property 'status' does not exist on type 'VideoImageValidationResult'.
);
}
}
Comment on lines +46 to +57

@codeant-ai codeant-ai Bot Jul 11, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: The new validation runs whenever an image URL is present, even when text content is already provided. In this flow OCR would not use the image, so valid text submissions can now fail incorrectly due to an unrelated image URL; gate validation to only run when OCR is actually needed. [incorrect condition logic]

Severity Level: Major ⚠️ 
- ❌ Text plus image submissions rejected for incidental bad image URL.
- ⚠️ Video generation fails when OCR would be skipped anyway.
Steps of Reproduction ✅ 
1. Issue a POST request to `/api/video/generate` (handled by `POST` in
`src/app/api/video/generate/route.ts:23`) with a JSON body containing both a valid
`content` string and an `imageUrl` pointing to a malformed or non-image payload (for
example a URL returning plain text, as modeled in
`src/__tests__/api/video-generate-route.test.ts:50-55`).

2. The request is parsed and validated by `parseAndValidateRequest(req,
generateVideoSchema)` in `src/app/api/video/generate/route.ts:43`, using
`generateVideoSchema` from `src/lib/validations/video.ts:4-9`, which explicitly allows
both `content` and `imageUrl` and only requires that at least one of them be present.

3. Because `data.imageUrl` is non-null, the condition at
`src/app/api/video/generate/route.ts:46` executes `validateVideoImageUrl(data.imageUrl)`
from `src/lib/video/image-validation.ts:49-82`, which performs a fetch and magic-byte
validation and returns `{ ok: false, status: 422, code: "INVALID_IMAGE_PAYLOAD", error:
... }` for malformed image payloads.

4. The handler then returns a 422 response at `src/app/api/video/generate/route.ts:48-55`
and never enqueues the `video/generate.requested` event for Inngest, even though the text
`content` alone is sufficient for `runVideoPipeline({ content, imageUrl })` in
`src/inngest/functions.ts:15-24` and `src/lib/video/pipeline.ts:102-120` (where OCR is
only run when `imageUrl` is present and `content` is absent), causing otherwise valid
text-based submissions with an incidental bad image URL to be rejected.

Fix in Cursor Fix in VSCode Claude

(Use Cmd/Ctrl + Click for best experience)

Prompt for AI Agent 🤖 
This is a comment left during a code review.

**Path:** src/app/api/video/generate/route.ts
**Line:** 46:57
**Comment:**
	*Incorrect Condition Logic: The new validation runs whenever an image URL is present, even when text content is already provided. In this flow OCR would not use the image, so valid text submissions can now fail incorrectly due to an unrelated image URL; gate validation to only run when OCR is actually needed.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
Once fix is implemented, also check other comments on the same PR, and ask user if the user wants to fix the rest of the comments as well. if said yes, then fetch all the comments validate the correctness and implement a minimal fix
👍 | 👎


// One active generation per user. The background job releases this lock when it
// finishes (success or failure); a 5-minute TTL guards against leaked locks.
const lockKey = `video_lock:${user.id}`;
Expand Down
82 changes: 82 additions & 0 deletions src/lib/video/image-validation.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,82 @@
const PNG_SIGNATURE = [0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a];
const JPEG_SIGNATURE = [0xff, 0xd8, 0xff];
const WEBP_SIGNATURE_PREFIX = [0x52, 0x49, 0x46, 0x46];
const WEBP_SIGNATURE_SUFFIX = [0x57, 0x45, 0x42, 0x50];

export const VIDEO_IMAGE_ALLOWED_MIME_TYPES = [
'image/png',
'image/jpeg',
'image/webp',
] as const;

export const VIDEO_IMAGE_ALLOWED_TYPES_LABEL = 'PNG, JPG, or WEBP';

export type VideoImageValidationResult =
| { ok: true; mimeType: string }
| {
ok: false;
status: 422 | 500;
code: string;
error: string;
};

export function isAllowedVideoImageMimeType(mimeType: string) {
return (VIDEO_IMAGE_ALLOWED_MIME_TYPES as readonly string[]).includes(
mimeType.toLowerCase(),
);
}

export function detectVideoImageMimeType(bytes: Uint8Array): string | null {
if (bytes.length >= PNG_SIGNATURE.length && PNG_SIGNATURE.every((byte, index) => bytes[index] === byte)) {
return 'image/png';
}

if (bytes.length >= JPEG_SIGNATURE.length && JPEG_SIGNATURE.every((byte, index) => bytes[index] === byte)) {
return 'image/jpeg';
}
Comment on lines +34 to +36

@codeant-ai codeant-ai Bot Jul 11, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: The JPEG detector accepts any 3-byte prefix FF D8 FF, which can classify invalid/truncated payloads as JPEG and let malformed data pass this gate. Strengthen JPEG validation with a longer signature check (and/or minimal structural checks) so malformed payloads fail fast as intended. [logic error]

Severity Level: Major ⚠️ 
- ⚠️ Malformed JPEG-like payloads can pass validation gate.
- ⚠️ Downstream image consumers may still fail unexpectedly.
Steps of Reproduction ✅ 
1. An attacker hosts an endpoint that returns a payload whose first three bytes match the
JPEG magic prefix `[0xff, 0xd8, 0xff]` but whose remaining bytes are arbitrary non-JPEG
data (for example, `0xff 0xd8 0xff` followed by text or random bytes).

2. A client sends a POST request to `/api/video/generate` (handled by `POST` in
`src/app/api/video/generate/route.ts:23`) with `imageUrl` pointing to this endpoint, which
passes `safeUrl` validation in `src/lib/validations/common.ts:3-4` via
`generateVideoSchema` in `src/lib/validations/video.ts:4-9`.

3. The route calls `validateVideoImageUrl(data.imageUrl)` at
`src/app/api/video/generate/route.ts:46-47`, which fetches the URL, converts the body to a
`Uint8Array` at `src/lib/video/image-validation.ts:61`, and then invokes
`detectVideoImageMimeType(bytes)` at `src/lib/video/image-validation.ts:62`.

4. Inside `detectVideoImageMimeType`, the JPEG branch at
`src/lib/video/image-validation.ts:34-36` only checks that `bytes.length >= 3` and that
the first three bytes equal the JPEG signature, so it returns `'image/jpeg'` even though
the payload is not structurally valid; `validateVideoImageUrl` then treats this as `ok:
true` at `src/lib/video/image-validation.ts:73`, allowing malformed data to pass the
validation gate instead of failing fast as intended.

Fix in Cursor Fix in VSCode Claude

(Use Cmd/Ctrl + Click for best experience)

Prompt for AI Agent 🤖 
This is a comment left during a code review.

**Path:** src/lib/video/image-validation.ts
**Line:** 34:36
**Comment:**
	*Logic Error: The JPEG detector accepts any 3-byte prefix `FF D8 FF`, which can classify invalid/truncated payloads as JPEG and let malformed data pass this gate. Strengthen JPEG validation with a longer signature check (and/or minimal structural checks) so malformed payloads fail fast as intended.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
Once fix is implemented, also check other comments on the same PR, and ask user if the user wants to fix the rest of the comments as well. if said yes, then fetch all the comments validate the correctness and implement a minimal fix
👍 | 👎


if (
bytes.length >= 12 &&
WEBP_SIGNATURE_PREFIX.every((byte, index) => bytes[index] === byte) &&
WEBP_SIGNATURE_SUFFIX.every((byte, index) => bytes[index + 8] === byte)
) {
return 'image/webp';
}

return null;
}

export async function validateVideoImageUrl(imageUrl: string): Promise<VideoImageValidationResult> {
try {
const response = await fetch(imageUrl, { cache: 'no-store' });

@codeant-ai codeant-ai Bot Jul 11, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: The server fetches a user-controlled URL directly, which enables SSRF against internal network targets (for example cloud metadata or private services). Restrict allowed hosts/IP ranges (block localhost, link-local, RFC1918, etc.) before fetching, and reject non-public destinations. [ssrf]

Severity Level: Critical 🚨 
- ❌ Video generation endpoint can query internal metadata services.
- ❌ Attackers can reach private network services via imageUrl.
- ⚠️ SSRF may bypass perimeter network protections entirely.
Steps of Reproduction ✅ 
1. An attacker sends a POST request to `/api/video/generate` (handled by `POST` in
`src/app/api/video/generate/route.ts:23`) with a JSON body where `imageUrl` is a URL to an
internal resource, such as `http://169.254.169.254/latest/meta-data/...`, and optionally
omits `content`.

2. The body is parsed by `parseAndValidateRequest` against `generateVideoSchema` in
`src/lib/validations/video.ts:4-9`, where `imageUrl` is validated only as a `safeUrl`
(`trimmedString.url().max(2048)` in `src/lib/validations/common.ts:3-4`) with no host or
IP-range restrictions, so internal network endpoints are accepted.

3. Because `data.imageUrl` is present, the route calls
`validateVideoImageUrl(data.imageUrl)` at `src/app/api/video/generate/route.ts:46-47`,
which in turn executes `fetch(imageUrl, { cache: 'no-store' })` at
`src/lib/video/image-validation.ts:51` directly against the attacker-controlled URL from
the server environment.

4. On infrastructure where the backend has network access to cloud metadata services or
other internal systems, this design allows the attacker to cause server-side requests to
those internal endpoints (SSRF) and observe success or failure via the validation response
from `validateVideoImageUrl`, potentially exfiltrating sensitive information or
interacting with private services.

Fix in Cursor Fix in VSCode Claude

(Use Cmd/Ctrl + Click for best experience)

Prompt for AI Agent 🤖 
This is a comment left during a code review.

**Path:** src/lib/video/image-validation.ts
**Line:** 51:51
**Comment:**
	*Ssrf: The server fetches a user-controlled URL directly, which enables SSRF against internal network targets (for example cloud metadata or private services). Restrict allowed hosts/IP ranges (block localhost, link-local, RFC1918, etc.) before fetching, and reject non-public destinations.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
Once fix is implemented, also check other comments on the same PR, and ask user if the user wants to fix the rest of the comments as well. if said yes, then fetch all the comments validate the correctness and implement a minimal fix
👍 | 👎

if (!response.ok) {
return {
ok: false,
status: 422,
code: 'INVALID_IMAGE_PAYLOAD',
error: `Please upload a valid ${VIDEO_IMAGE_ALLOWED_TYPES_LABEL} image.`,
};
}

const bytes = new Uint8Array(await response.arrayBuffer());

@codeant-ai codeant-ai Bot Jul 11, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: This reads the entire remote response into memory with no size cap, so a very large payload can exhaust memory and crash or degrade the server. Enforce a strict maximum byte limit (via Content-Length check plus streamed read cap) and fail once the cap is exceeded. [performance]

Severity Level: Major ⚠️ 
- ❌ Large image responses can crash video generation backend.
- ⚠️ Memory pressure degrades performance for legitimate video jobs.
Steps of Reproduction ✅ 
1. An attacker hosts a server at a public URL and configures it to return a very large
HTTP response body (for example, hundreds of megabytes or several gigabytes) with status
200 when requested.

2. The attacker sends a POST request to `/api/video/generate` (handled by `POST` in
`src/app/api/video/generate/route.ts:23`) with `imageUrl` pointing to this large-response
endpoint and no `content`, which is accepted because `imageUrl` passes the `safeUrl`
validation in `src/lib/validations/common.ts:3-4` via `generateVideoSchema` in
`src/lib/validations/video.ts:4-9`.

3. The route calls `validateVideoImageUrl(data.imageUrl)` at
`src/app/api/video/generate/route.ts:46-47`, which fetches the URL and then executes
`const bytes = new Uint8Array(await response.arrayBuffer());` at
`src/lib/video/image-validation.ts:61`, causing the entire remote response body to be
buffered into memory with no explicit size cap.

4. Under sufficiently large responses or multiple concurrent malicious requests, this
unbounded allocation can exhaust the server process’s memory or trigger heavy garbage
collection, degrading performance or crashing the video generation API and its associated
background processing.

Fix in Cursor Fix in VSCode Claude

(Use Cmd/Ctrl + Click for best experience)

Prompt for AI Agent 🤖 
This is a comment left during a code review.

**Path:** src/lib/video/image-validation.ts
**Line:** 61:61
**Comment:**
	*Performance: This reads the entire remote response into memory with no size cap, so a very large payload can exhaust memory and crash or degrade the server. Enforce a strict maximum byte limit (via Content-Length check plus streamed read cap) and fail once the cap is exceeded.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
Once fix is implemented, also check other comments on the same PR, and ask user if the user wants to fix the rest of the comments as well. if said yes, then fetch all the comments validate the correctness and implement a minimal fix
👍 | 👎

const mimeType = detectVideoImageMimeType(bytes);

if (!mimeType || !isAllowedVideoImageMimeType(mimeType)) {
return {
ok: false,
status: 422,
code: 'INVALID_IMAGE_PAYLOAD',
error: `Please upload a valid ${VIDEO_IMAGE_ALLOWED_TYPES_LABEL} image.`,
};
}

return { ok: true, mimeType };
} catch {
return {
ok: false,
status: 422,
code: 'INVALID_IMAGE_PAYLOAD',
error: `Please upload a valid ${VIDEO_IMAGE_ALLOWED_TYPES_LABEL} image.`,
};
}
}
Loading