@effectnode/media 0.1.0

package/dist/backend/movie-backend/agent/storyprompt.d.ts

@@ -0,0 +1 @@
+export declare const storyPrompt = "# Vid Script Builder\n\n## Role & Purpose\n\nYou are a short-form video planner and AI Prompt Specialist specializing in humorous, viral content. Your mission is to take user-provided video topics, comedy concepts, or scenarios and convert them directly into structured CSV data designed for Text-to-Image (T2I) and Image-to-Video (I2V) production workflows. Each row represents a single moment of a camera shot, with its own duration (3\u201320 seconds).\n\n## Workflow & Output Rules\n\nWhen a user provides a theme, scenario, or prompt request, adhere strictly to the following specifications.\n\n### 1. Table Schema (CSV Format)\n\nEvery output must include the following four columns:\n\n| Column     | Description                               |\n| ---------- | ----------------------------------------- |\n| `id`       | Sequential moment number (1, 2, 3...)     |\n| `duration` | Duration of this moment in seconds (3\u201320) |\n| `t2i`      | Text-to-Image Prompt (English)            |\n| `i2v`      | Image-to-Video Prompt (English)           |\n\n### 2. Scene Breakdown & Segmentation Constraint\n\n- **Each item is a single moment of a camera shot.** Treat every row as one discrete moment/beat within a shot \u2014 not a whole scene.\n- **Break large scenes into many small moments.** When a user provides a broad scene, storyline, or concept, decompose it into as many single-moment rows as possible \u2014 never compress a large scene into one row.\n- Each row carries a `duration` between **3 and 20 seconds**, but **prefer small moments**: break the action into the smallest coherent beats and favor short durations (3\u20136 seconds); use a longer duration only when one continuous camera move or action genuinely requires it.\n- Each moment captures exactly one action beat, one camera setup, and one comedic/punchline step.\n- Maintain narrative continuity across consecutive moments (consistent characters, setting, and progression) so the sequence edits together into one coherent video.\n- Pacing, narrative action, character dialogue, and voiceover in each moment must be calibrated to fit strictly within its stated duration.\n\n### 3. t2i (Text-to-Image) Guidelines \u2014 Z-Image-Turbo\n\n- **Language:** English.\n- **Model:** Optimized for **Z-Image-Turbo**. Write prompts as **natural-language sentences**, never comma-separated keyword tags (`(masterpiece), 8k, trending on artstation` fails).\n- **Subject first:** Lead with the subject and its action \u2014 Z-Image-Turbo weights earlier tokens more heavily, so never bury the main subject late in the description.\n- **Character consistency (critical across shots):** before generating any prompts, write one fixed, detailed character description (a \"character bible\") for every recurring character \u2014 name, age, face shape, hair color/style, build, outfit, and any distinctive feature (scar, glasses, earring). Repeat that exact description verbatim at the start of every t2i prompt featuring that character, varying only the shot-specific action, expression, and pose. Never reword or omit core appearance details between shots, or the character's identity will drift across scenes.\n- **One focused paragraph** built in this layered order:\n  1. **Subject & Action** \u2014 who/what, doing what, with specific physical details.\n  2. **Environment & Context** \u2014 location, time of day, weather, background.\n  3. **Lighting & Atmosphere** \u2014 the strongest lever after style.\n  4. **Visual Style** \u2014 pick **one** primary style (e.g., cinematic photorealistic) plus camera/lens/film details.\n  5. **Composition** \u2014 framing, focus, close-up vs. wide shot.\n- **No negative prompts:** Z-Image-Turbo has no classifier-free guidance, so negative prompts are unsupported. Phrase constraints positively (e.g., \"a quiet, deserted street\" instead of \"no people\").\n- **Be specific, avoid vagueness:** replace \"a dog\" with \"a golden retriever puppy with one ear flopped sideways, sitting on a porch step.\"\n- **Keep quality modifiers minimal:** 2\u20133 technical tags max (e.g., \"85mm f/1.8, shallow depth of field\").\n- **On-image text (if any):** wrap literal text in straight quotes, keep it to a short phrase, and specify font style and placement.\n- **Scope:** each prompt must describe a single moment of a camera shot, matching its stated duration.\n\n### 4. i2v (Image-to-Video) Guidelines \u2014 LTX-2.3\n\n- **Language:** English.\n- **Core rule \u2014 describe the motion, not the image.** The t2i image is the anchor that already fixes the subject, setting, and lighting. Do not re-describe static details; write only what changes: motion, camera movement, and performance.\n- **Motion-first:** lead with action verbs and camera verbs in the first 20\u201340 words. Describe \"how the pixels move,\" not the subject itself.\n- **Scale motion to the moment's duration:** one main action beat plus one simple camera move for short moments (3\u20136s); add a second beat or a longer continuous move only when the duration exceeds ~10s. Do not stack simultaneous motions, and avoid complex camera terms (e.g., \"rack focus\", \"dolly zoom\") that the distilled model drifts on.\n- **Avoid:** generic phrasing (\"make it look nice\"), contradictory directions (\"fast dramatic zoom but calm and slow\"), still-photo phrasing (\"preserve identity\", \"gentle parallax\"), and prompts over ~150 words.\n- **Never** include the clip duration or the model name in the prompt text.\n- **Negative prompt:** keep negatives in the dedicated negative field (e.g., `no stiff motion, no flat lighting, no artifacts, no morphing, no flickering`), never inside the positive prompt.\n\nEvery i2v prompt must still explicitly include these four components, written as a single cinematic paragraph:\n\n| #   | Component           | Description                                                                               |\n| --- | ------------------- | ----------------------------------------------------------------------------------------- |\n| 1   | Camera Movement     | One specific motion directive (e.g., slow dolly-in, steady hold, snap zoom).              |\n| 2   | Character Name      | Named subject performing the action \u2014 name once, then focus on their motion/performance.  |\n| 3   | Character Dialogue  | Spoken lines in quotation marks, treated as an audio/lipsync cue.                         |\n| 4   | Voiceover/Narration | Formatted as `Voiceover: \"...\"` to capture the punchline, inner monologue, or core humor. |\n\n## Output Format Standard\n\nOutput the raw content inside a clean CSV code block, followed immediately by a rendered Markdown preview table.\n\n```csv\nid,duration,t2i,i2v\n1,5,\"[English T2I Prompt]\",\"[English I2V Prompt: Includes camera movement, character name, dialogue, and voiceover narration structured for this moment's duration]\"\n```\n\n### Sample Character Bible (reused across every shot)\n\n> Alex \u2014 a 20-year-old male college student with short black hair, round wire-frame glasses, and a faded blue hoodie, always with a nervous, wide-eyed look. Repeat this exact description verbatim at the start of every t2i prompt that features Alex, changing only his action and expression per moment.\n\n### Sample t2i Prompt Structure (single moment)\n\n> A close-up of Alex, a college student with wide, panicked eyes and sweat beading on his forehead, frantically gripping a pen in an exam hall. Harsh fluorescent overhead lighting, rows of wooden desks receding into the background, other students blurred in soft focus. Cinematic photorealistic style, shot on a 50mm lens at f/2.0, shallow depth of field, tense anxious atmosphere.\n\n### Sample i2v Prompt Structure (single moment)\n\n> Slow dolly-in toward Alex's face as his eyes widen and a bead of sweat rolls down his temple. He grips the pen tighter and whispers urgently: \"I'm so dead, I literally haven't opened the textbook!\" Voiceover: \"Every class has that one top student whose catchphrase is 'I failed,' right before scoring 100%.\"\n\n> **Negative prompt:** `no stiff motion, no flat lighting, no artifacts, no morphing, no flickering`\n";

package/dist/backend/movie-backend/agent/storyprompt.js

@@ -0,0 +1,90 @@
+export const storyPrompt = `# Vid Script Builder
+
+## Role & Purpose
+
+You are a short-form video planner and AI Prompt Specialist specializing in humorous, viral content. Your mission is to take user-provided video topics, comedy concepts, or scenarios and convert them directly into structured CSV data designed for Text-to-Image (T2I) and Image-to-Video (I2V) production workflows. Each row represents a single moment of a camera shot, with its own duration (3–20 seconds).
+
+## Workflow & Output Rules
+
+When a user provides a theme, scenario, or prompt request, adhere strictly to the following specifications.
+
+### 1. Table Schema (CSV Format)
+
+Every output must include the following four columns:
+
+| Column     | Description                               |
+| ---------- | ----------------------------------------- |
+| \`id\`       | Sequential moment number (1, 2, 3...)     |
+| \`duration\` | Duration of this moment in seconds (3–20) |
+| \`t2i\`      | Text-to-Image Prompt (English)            |
+| \`i2v\`      | Image-to-Video Prompt (English)           |
+
+### 2. Scene Breakdown & Segmentation Constraint
+
+- **Each item is a single moment of a camera shot.** Treat every row as one discrete moment/beat within a shot — not a whole scene.
+- **Break large scenes into many small moments.** When a user provides a broad scene, storyline, or concept, decompose it into as many single-moment rows as possible — never compress a large scene into one row.
+- Each row carries a \`duration\` between **3 and 20 seconds**, but **prefer small moments**: break the action into the smallest coherent beats and favor short durations (3–6 seconds); use a longer duration only when one continuous camera move or action genuinely requires it.
+- Each moment captures exactly one action beat, one camera setup, and one comedic/punchline step.
+- Maintain narrative continuity across consecutive moments (consistent characters, setting, and progression) so the sequence edits together into one coherent video.
+- Pacing, narrative action, character dialogue, and voiceover in each moment must be calibrated to fit strictly within its stated duration.
+
+### 3. t2i (Text-to-Image) Guidelines — Z-Image-Turbo
+
+- **Language:** English.
+- **Model:** Optimized for **Z-Image-Turbo**. Write prompts as **natural-language sentences**, never comma-separated keyword tags (\`(masterpiece), 8k, trending on artstation\` fails).
+- **Subject first:** Lead with the subject and its action — Z-Image-Turbo weights earlier tokens more heavily, so never bury the main subject late in the description.
+- **Character consistency (critical across shots):** before generating any prompts, write one fixed, detailed character description (a "character bible") for every recurring character — name, age, face shape, hair color/style, build, outfit, and any distinctive feature (scar, glasses, earring). Repeat that exact description verbatim at the start of every t2i prompt featuring that character, varying only the shot-specific action, expression, and pose. Never reword or omit core appearance details between shots, or the character's identity will drift across scenes.
+- **One focused paragraph** built in this layered order:
+  1. **Subject & Action** — who/what, doing what, with specific physical details.
+  2. **Environment & Context** — location, time of day, weather, background.
+  3. **Lighting & Atmosphere** — the strongest lever after style.
+  4. **Visual Style** — pick **one** primary style (e.g., cinematic photorealistic) plus camera/lens/film details.
+  5. **Composition** — framing, focus, close-up vs. wide shot.
+- **No negative prompts:** Z-Image-Turbo has no classifier-free guidance, so negative prompts are unsupported. Phrase constraints positively (e.g., "a quiet, deserted street" instead of "no people").
+- **Be specific, avoid vagueness:** replace "a dog" with "a golden retriever puppy with one ear flopped sideways, sitting on a porch step."
+- **Keep quality modifiers minimal:** 2–3 technical tags max (e.g., "85mm f/1.8, shallow depth of field").
+- **On-image text (if any):** wrap literal text in straight quotes, keep it to a short phrase, and specify font style and placement.
+- **Scope:** each prompt must describe a single moment of a camera shot, matching its stated duration.
+
+### 4. i2v (Image-to-Video) Guidelines — LTX-2.3
+
+- **Language:** English.
+- **Core rule — describe the motion, not the image.** The t2i image is the anchor that already fixes the subject, setting, and lighting. Do not re-describe static details; write only what changes: motion, camera movement, and performance.
+- **Motion-first:** lead with action verbs and camera verbs in the first 20–40 words. Describe "how the pixels move," not the subject itself.
+- **Scale motion to the moment's duration:** one main action beat plus one simple camera move for short moments (3–6s); add a second beat or a longer continuous move only when the duration exceeds ~10s. Do not stack simultaneous motions, and avoid complex camera terms (e.g., "rack focus", "dolly zoom") that the distilled model drifts on.
+- **Avoid:** generic phrasing ("make it look nice"), contradictory directions ("fast dramatic zoom but calm and slow"), still-photo phrasing ("preserve identity", "gentle parallax"), and prompts over ~150 words.
+- **Never** include the clip duration or the model name in the prompt text.
+- **Negative prompt:** keep negatives in the dedicated negative field (e.g., \`no stiff motion, no flat lighting, no artifacts, no morphing, no flickering\`), never inside the positive prompt.
+
+Every i2v prompt must still explicitly include these four components, written as a single cinematic paragraph:
+
+| #   | Component           | Description                                                                               |
+| --- | ------------------- | ----------------------------------------------------------------------------------------- |
+| 1   | Camera Movement     | One specific motion directive (e.g., slow dolly-in, steady hold, snap zoom).              |
+| 2   | Character Name      | Named subject performing the action — name once, then focus on their motion/performance.  |
+| 3   | Character Dialogue  | Spoken lines in quotation marks, treated as an audio/lipsync cue.                         |
+| 4   | Voiceover/Narration | Formatted as \`Voiceover: "..."\` to capture the punchline, inner monologue, or core humor. |
+
+## Output Format Standard
+
+Output the raw content inside a clean CSV code block, followed immediately by a rendered Markdown preview table.
+
+\`\`\`csv
+id,duration,t2i,i2v
+1,5,"[English T2I Prompt]","[English I2V Prompt: Includes camera movement, character name, dialogue, and voiceover narration structured for this moment's duration]"
+\`\`\`
+
+### Sample Character Bible (reused across every shot)
+
+> Alex — a 20-year-old male college student with short black hair, round wire-frame glasses, and a faded blue hoodie, always with a nervous, wide-eyed look. Repeat this exact description verbatim at the start of every t2i prompt that features Alex, changing only his action and expression per moment.
+
+### Sample t2i Prompt Structure (single moment)
+
+> A close-up of Alex, a college student with wide, panicked eyes and sweat beading on his forehead, frantically gripping a pen in an exam hall. Harsh fluorescent overhead lighting, rows of wooden desks receding into the background, other students blurred in soft focus. Cinematic photorealistic style, shot on a 50mm lens at f/2.0, shallow depth of field, tense anxious atmosphere.
+
+### Sample i2v Prompt Structure (single moment)
+
+> Slow dolly-in toward Alex's face as his eyes widen and a bead of sweat rolls down his temple. He grips the pen tighter and whispers urgently: "I'm so dead, I literally haven't opened the textbook!" Voiceover: "Every class has that one top student whose catchphrase is 'I failed,' right before scoring 100%."
+
+> **Negative prompt:** \`no stiff motion, no flat lighting, no artifacts, no morphing, no flickering\`
+`;

package/dist/backend/movie-backend/agent/tools/edit-image.d.ts

@@ -0,0 +1,3 @@
+import type { AgentTool } from "./types";
+declare const tool: AgentTool;
+export default tool;

package/dist/backend/movie-backend/agent/tools/edit-image.js

@@ -0,0 +1,192 @@
+import { existsSync, mkdirSync, readFileSync, writeFileSync, realpathSync, } from "node:fs";
+import { join, sep } from "node:path";
+import { isValidProjectId, resolveWorkspacePath, classifyFile, AGENT_UPLOAD_DIR, workspaceDir, } from "../workspace";
+import sharp from "sharp";
+/** Default output resolution (px) when the tool is called without one. */
+const DEFAULT_RESOLUTION = 512;
+/** Allowed resolution values (px). */
+const ALLOWED_RESOLUTIONS = [512, 1024];
+/** Resize an image to at most `maxDim` on the longest edge, keeping aspect ratio. */
+async function resizeToMax(input, maxDim) {
+    const resized = await sharp(input)
+        .resize({
+        width: maxDim,
+        height: maxDim,
+        fit: "inside",
+        withoutEnlargement: true,
+    })
+        .png()
+        .toBuffer();
+    const meta = await sharp(resized).metadata();
+    return {
+        buffer: resized,
+        width: meta.width ?? maxDim,
+        height: meta.height ?? maxDim,
+    };
+}
+async function readSSE(response, onEvent) {
+    const reader = response.body?.getReader();
+    if (!reader)
+        return;
+    const decoder = new TextDecoder();
+    let buffer = "";
+    try {
+        while (true) {
+            const { done, value } = await reader.read();
+            if (done)
+                break;
+            buffer += decoder.decode(value, { stream: true });
+            const lines = buffer.split("\n");
+            buffer = lines.pop() || "";
+            let eventType = "message";
+            for (const line of lines) {
+                if (line.startsWith("event: ")) {
+                    eventType = line.slice(7).trim();
+                }
+                else if (line.startsWith("data: ")) {
+                    try {
+                        onEvent(eventType, JSON.parse(line.slice(6)));
+                    }
+                    catch {
+                        // skip malformed lines
+                    }
+                    eventType = "message";
+                }
+            }
+        }
+    }
+    finally {
+        reader.releaseLock();
+    }
+}
+const tool = {
+    name: "edit_image",
+    description: "Edit an image in the workspace using a text prompt (e.g. change the background, style, or expression).",
+    parameters: {
+        type: "object",
+        properties: {
+            prompt: { type: "string", description: "The edit instruction" },
+            image: {
+                type: "string",
+                description: "Path of the image in the workspace to edit",
+            },
+            resolution: {
+                type: "number",
+                enum: [512, 1024],
+                description: "Output resolution in pixels (512 or 1024, default 512)",
+            },
+        },
+        required: ["prompt", "image"],
+    },
+    run: async (args, ctx) => {
+        if (!isValidProjectId(ctx.projectId))
+            return "Invalid project ID.";
+        const prompt = typeof args.prompt === "string" ? args.prompt.trim() : "";
+        if (!prompt)
+            return "edit_image requires a prompt.";
+        const image = typeof args.image === "string" ? args.image.trim() : "";
+        if (!image)
+            return "edit_image requires an image from the workspace.";
+        const resolution = typeof args.resolution === "number" &&
+            ALLOWED_RESOLUTIONS.includes(args.resolution)
+            ? args.resolution
+            : DEFAULT_RESOLUTION;
+        const abs = resolveWorkspacePath(ctx.projectId, image);
+        if (!abs || !existsSync(abs)) {
+            return `Image not found: ${image}`;
+        }
+        // Classify the resolved real path (resolves symlinks) so a symlink named
+        // like an image but pointing at another file is not treated as an image.
+        let realAbs;
+        try {
+            realAbs = realpathSync(abs);
+        }
+        catch {
+            return `Image not found: ${image}`;
+        }
+        // Confirm the resolved (symlink-followed) path is still inside the
+        // project's workspace before reading it.
+        const realBase = realpathSync(workspaceDir(ctx.projectId));
+        if (realAbs !== realBase && !realAbs.startsWith(realBase + sep)) {
+            return `Image outside workspace: ${image}`;
+        }
+        if (classifyFile(realAbs) !== "image") {
+            return `Not an image: ${image}`;
+        }
+        if (!ctx.backendPort)
+            return "Image backend not available.";
+        // Resize the source image to at most the chosen resolution (keeping aspect
+        // ratio) and re-encode as PNG, then stage it into the project's
+        // agent-upload dir so the mlxgen endpoint can resolve it by bare filename.
+        let resized;
+        try {
+            resized = await resizeToMax(readFileSync(realAbs), resolution);
+        }
+        catch (e) {
+            return `Failed to process image: ${String(e)}`;
+        }
+        const uploadProjectDir = join(AGENT_UPLOAD_DIR, ctx.projectId);
+        if (!existsSync(uploadProjectDir)) {
+            mkdirSync(uploadProjectDir, { recursive: true });
+        }
+        const origBase = (image.split("/").pop() || "image.png").replace(/\.[^.]+$/, "");
+        const imageName = `agent-${Date.now()}-${origBase.replace(/[^a-zA-Z0-9._-]/g, "_")}.png`;
+        writeFileSync(join(uploadProjectDir, imageName), resized.buffer);
+        ctx.emit?.("notice", { text: `Editing image "${image}"...` });
+        try {
+            const body = {
+                prompt,
+                imagePath: imageName,
+                projectId: ctx.projectId,
+                width: resized.width,
+                height: resized.height,
+            };
+            const res = await fetch(`http://localhost:${ctx.backendPort}/api/mlxgen/generate`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify(body),
+                signal: ctx.signal,
+            });
+            if (!res.ok) {
+                return `Image edit failed: ${await res.text()}`;
+            }
+            let resultPath = null;
+            let resultFilename = null;
+            let error = null;
+            await readSSE(res, (event, data) => {
+                switch (event) {
+                    case "log":
+                        ctx.emit?.("notice", { text: data.text });
+                        break;
+                    case "complete":
+                        resultPath = data.path;
+                        resultFilename = data.filename;
+                        break;
+                    case "error":
+                        error = data.error || "Image edit failed";
+                        break;
+                }
+            });
+            if (error)
+                return error;
+            if (!resultPath)
+                return "Image edit did not produce a result.";
+            // Copy the edited image into the workspace so it is previewable and
+            // usable in subsequent steps.
+            const outName = resultFilename ||
+                resultPath.split("/").pop() ||
+                "edited.png";
+            const outDir = workspaceDir(ctx.projectId);
+            if (!existsSync(outDir))
+                mkdirSync(outDir, { recursive: true });
+            writeFileSync(join(outDir, outName), readFileSync(resultPath));
+            const url = `/api/agent/file/preview?projectId=${encodeURIComponent(ctx.projectId)}&path=${encodeURIComponent(outName)}`;
+            ctx.emit?.("image", { url, path: outName });
+            return `Edited image saved to "${outName}".`;
+        }
+        catch (e) {
+            return `Image edit error: ${String(e)}`;
+        }
+    },
+};
+export default tool;

package/dist/backend/movie-backend/agent/tools/get-time.d.ts

@@ -0,0 +1,3 @@
+import type { AgentTool } from "./types";
+declare const tool: AgentTool;
+export default tool;

package/dist/backend/movie-backend/agent/tools/get-time.js

@@ -0,0 +1,7 @@
+const tool = {
+    name: "get_time",
+    description: "Return the current date and time.",
+    parameters: { type: "object", properties: {}, additionalProperties: false },
+    run: () => new Date().toString(),
+};
+export default tool;

package/dist/backend/movie-backend/agent/tools/grep-files.d.ts

@@ -0,0 +1,3 @@
+import type { AgentTool } from "./types";
+declare const tool: AgentTool;
+export default tool;

package/dist/backend/movie-backend/agent/tools/grep-files.js

@@ -0,0 +1,59 @@
+import { existsSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+import { resolveWorkspacePath, workspaceDir, walkFiles } from "../workspace";
+const tool = {
+    name: "grep_files",
+    description: "Search for a regex pattern within text files in the workspace.",
+    parameters: {
+        type: "object",
+        properties: {
+            pattern: { type: "string", description: "Regular expression to search for" },
+            path: {
+                type: "string",
+                description: "Optional directory to search (defaults to the whole workspace)",
+            },
+        },
+        required: ["pattern"],
+    },
+    run: (args, ctx) => {
+        const pattern = typeof args.pattern === "string" ? args.pattern : "";
+        if (!pattern)
+            return "grep_files requires a pattern.";
+        let re;
+        try {
+            re = new RegExp(pattern, "i");
+        }
+        catch {
+            return `Invalid regex: ${pattern}`;
+        }
+        const dirArg = typeof args.path === "string" && args.path.trim() ? args.path.trim() : "";
+        const base = dirArg
+            ? resolveWorkspacePath(ctx.projectId, dirArg)
+            : workspaceDir(ctx.projectId);
+        if (!base || !existsSync(base)) {
+            return `Directory not found: ${dirArg || "(workspace)"}`;
+        }
+        const results = [];
+        for (const f of walkFiles(base)) {
+            if (f.kind !== "text")
+                continue;
+            const abs = join(base, f.path);
+            let content;
+            try {
+                content = readFileSync(abs, "utf-8");
+            }
+            catch {
+                continue;
+            }
+            content.split("\n").forEach((line, i) => {
+                if (re.test(line)) {
+                    results.push(`${f.path}:${i + 1}: ${line.trim()}`);
+                }
+            });
+        }
+        return results.length > 0
+            ? results.join("\n")
+            : `No matches for "${pattern}".`;
+    },
+};
+export default tool;

package/dist/backend/movie-backend/agent/tools/image-to-video-generation.d.ts

@@ -0,0 +1,3 @@
+import type { AgentTool } from "./types";
+declare const tool: AgentTool;
+export default tool;

package/dist/backend/movie-backend/agent/tools/image-to-video-generation.js

@@ -0,0 +1,160 @@
+import { existsSync, mkdirSync, readFileSync, writeFileSync, realpathSync, } from "node:fs";
+import { join } from "node:path";
+import { isValidProjectId, resolveWorkspacePath, classifyFile, AGENT_UPLOAD_DIR, workspaceDir, } from "../workspace";
+async function readSSE(response, onEvent) {
+    const reader = response.body?.getReader();
+    if (!reader)
+        return;
+    const decoder = new TextDecoder();
+    let buffer = "";
+    try {
+        while (true) {
+            const { done, value } = await reader.read();
+            if (done)
+                break;
+            buffer += decoder.decode(value, { stream: true });
+            const lines = buffer.split("\n");
+            buffer = lines.pop() || "";
+            let eventType = "message";
+            for (const line of lines) {
+                if (line.startsWith("event: ")) {
+                    eventType = line.slice(7).trim();
+                }
+                else if (line.startsWith("data: ")) {
+                    try {
+                        onEvent(eventType, JSON.parse(line.slice(6)));
+                    }
+                    catch {
+                        // skip malformed lines
+                    }
+                    eventType = "message";
+                }
+            }
+        }
+    }
+    finally {
+        reader.releaseLock();
+    }
+}
+const tool = {
+    name: "image_to_video_generation",
+    description: "Generate a video from a prompt and an image in the workspace.",
+    parameters: {
+        type: "object",
+        properties: {
+            prompt: { type: "string", description: "The video prompt" },
+            image: {
+                type: "string",
+                description: "Path of the image in the workspace to animate",
+            },
+            duration: {
+                type: "number",
+                description: "Video duration in seconds (default 5)",
+            },
+            mode: {
+                type: "string",
+                enum: ["distilled", "one-stage", "two-stage"],
+                description: "Generation mode: distilled (default), one-stage, or two-stage",
+            },
+        },
+        required: ["prompt", "image"],
+    },
+    run: async (args, ctx) => {
+        if (!isValidProjectId(ctx.projectId))
+            return "Invalid project ID.";
+        const prompt = typeof args.prompt === "string" ? args.prompt.trim() : "";
+        if (!prompt)
+            return "image_to_video_generation requires a prompt.";
+        const image = typeof args.image === "string" ? args.image.trim() : "";
+        if (!image)
+            return "image_to_video_generation requires an image from the workspace.";
+        const abs = resolveWorkspacePath(ctx.projectId, image);
+        if (!abs || !existsSync(abs)) {
+            return `Image not found: ${image}`;
+        }
+        // Classify the resolved real path (resolves symlinks) so a symlink named
+        // like an image but pointing at another file is not treated as an image.
+        let realAbs;
+        try {
+            realAbs = realpathSync(abs);
+        }
+        catch {
+            return `Image not found: ${image}`;
+        }
+        if (classifyFile(realAbs) !== "image") {
+            return `Not an image: ${image}`;
+        }
+        if (!ctx.backendPort)
+            return "Video backend not available.";
+        const duration = typeof args.duration === "number" && args.duration > 0
+            ? args.duration
+            : 5;
+        const frames = Math.round(duration * 24 + 1);
+        // Copy the image into the project's agent-upload dir so the video endpoint
+        // can resolve it by bare filename.
+        const uploadProjectDir = join(AGENT_UPLOAD_DIR, ctx.projectId);
+        if (!existsSync(uploadProjectDir)) {
+            mkdirSync(uploadProjectDir, { recursive: true });
+        }
+        const origName = image.split("/").pop() || "image.png";
+        const imageName = `agent-${Date.now()}-${origName.replace(/[^a-zA-Z0-9._-]/g, "_")}`;
+        writeFileSync(join(uploadProjectDir, imageName), readFileSync(realAbs));
+        ctx.emit?.("notice", { text: `Generating video from "${image}"...` });
+        try {
+            const body = {
+                prompt,
+                imagePath: imageName,
+                projectId: ctx.projectId,
+                outputDir: workspaceDir(ctx.projectId),
+                width: 448,
+                height: 448,
+                frames,
+                frameRate: 24,
+            };
+            if (typeof args.mode === "string" &&
+                ["distilled", "one-stage", "two-stage"].includes(args.mode)) {
+                body.mode = args.mode;
+            }
+            const res = await fetch(`http://localhost:${ctx.backendPort}/api/render/image-to-video`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify(body),
+                signal: ctx.signal,
+            });
+            if (!res.ok) {
+                return `Video generation failed: ${await res.text()}`;
+            }
+            let resultPath = null;
+            let resultFilename = null;
+            let error = null;
+            await readSSE(res, (event, data) => {
+                switch (event) {
+                    case "log":
+                        ctx.emit?.("notice", { text: data.text });
+                        break;
+                    case "complete":
+                        resultPath = data.path;
+                        resultFilename = data.filename;
+                        break;
+                    case "error":
+                        error = data.error || "Video generation failed";
+                        break;
+                }
+            });
+            if (error)
+                return error;
+            if (!resultPath)
+                return "Video generation did not produce a result.";
+            const relPath = resultFilename ||
+                resultPath.split("/").pop() ||
+                "video.mp4";
+            const url = `/api/agent/file/preview?projectId=${encodeURIComponent(ctx.projectId)}&path=${encodeURIComponent(relPath)}`;
+            ctx.emit?.("video", { url });
+            return `Generated video from "${image}": ${resultPath}`;
+        }
+        catch (e) {
+            return `Video generation error: ${String(e)}`;
+        }
+    },
+};
+export default tool;

package/dist/backend/movie-backend/agent/tools/index.d.ts

@@ -0,0 +1,7 @@
+import type { AgentTool, ToolRunContext } from "./types";
+export type { AgentTool, ToolRunContext } from "./types";
+export declare const TOOLS: AgentTool[];
+/** Build the OpenAI `tools` array from the tool objects. */
+export declare function toolDefinitions(tools: AgentTool[]): any[];
+/** Dispatch a tool call to the matching tool's `run` function. */
+export declare function runTool(tools: AgentTool[], name: string, args: string, ctx: ToolRunContext): Promise<string>;

package/dist/backend/movie-backend/agent/tools/index.js

@@ -0,0 +1,59 @@
+import getTimeTool from "./get-time";
+import listFilesTool from "./list-files";
+import readFileTool from "./read-file";
+import writeFileTool from "./write-file";
+import updateFileTool from "./update-file";
+import removeFileTool from "./remove-file";
+import renameFileTool from "./rename-file";
+import grepFilesTool from "./grep-files";
+import showImageTool from "./show-image";
+import editImageTool from "./edit-image";
+import imageToVideoGenerationTool from "./image-to-video-generation";
+import textToVideoGenerationTool from "./text-to-video-generation";
+import stitchVideosTool from "./stitch-videos";
+export const TOOLS = [
+    getTimeTool,
+    listFilesTool,
+    readFileTool,
+    writeFileTool,
+    updateFileTool,
+    removeFileTool,
+    renameFileTool,
+    grepFilesTool,
+    showImageTool,
+    editImageTool,
+    imageToVideoGenerationTool,
+    textToVideoGenerationTool,
+    stitchVideosTool,
+];
+/** Build the OpenAI `tools` array from the tool objects. */
+export function toolDefinitions(tools) {
+    return tools.map((t) => ({
+        type: "function",
+        function: {
+            name: t.name,
+            description: t.description,
+            parameters: t.parameters,
+        },
+    }));
+}
+/** Dispatch a tool call to the matching tool's `run` function. */
+export async function runTool(tools, name, args, ctx) {
+    const tool = tools.find((t) => t.name === name);
+    if (!tool)
+        return `Unknown tool: ${name}`;
+    let parsed = {};
+    try {
+        parsed = args ? JSON.parse(args) : {};
+    }
+    catch {
+        // leave empty on malformed args
+    }
+    try {
+        const out = await tool.run(parsed, ctx);
+        return typeof out === "string" ? out : String(out);
+    }
+    catch (e) {
+        return `Tool error: ${String(e)}`;
+    }
+}

package/dist/backend/movie-backend/agent/tools/list-files.d.ts

@@ -0,0 +1,3 @@
+import type { AgentTool } from "./types";
+declare const tool: AgentTool;
+export default tool;

package/dist/backend/movie-backend/agent/tools/list-files.js

@@ -0,0 +1,17 @@
+import { existsSync } from "node:fs";
+import { workspaceDir, walkFiles } from "../workspace";
+const tool = {
+    name: "list_files",
+    description: "List all files in the agent's workspace (recursive).",
+    parameters: { type: "object", properties: {}, additionalProperties: false },
+    run: (_args, ctx) => {
+        const base = workspaceDir(ctx.projectId);
+        if (!existsSync(base))
+            return "Workspace is empty.";
+        const files = walkFiles(base);
+        if (files.length === 0)
+            return "Workspace is empty.";
+        return JSON.stringify(files.map((f) => ({ path: f.path, kind: f.kind, size: f.size })));
+    },
+};
+export default tool;

package/dist/backend/movie-backend/agent/tools/read-file.d.ts

@@ -0,0 +1,3 @@
+import type { AgentTool } from "./types";
+declare const tool: AgentTool;
+export default tool;