vargai 0.4.0-alpha106 → 0.4.0-alpha108

package/package.json

@@ -28,7 +28,6 @@
     "@commitlint/config-conventional": "^20.0.0",
     "@size-limit/preset-small-lib": "^11.2.0",
     "@types/bun": "latest",
-    "@types/opentype.js": "^1.3.9",
     "@types/react": "^19.2.7",
     "husky": "^9.1.7",
     "lint-staged": "^16.2.7"
@@ -59,11 +58,9 @@
     "ai": "^6.0.26",
     "apify-client": "^2.20.0",
     "citty": "^0.1.6",
-    "fflate": "^0.8.2",
     "fluent-ffmpeg": "^2.1.3",
     "groq-sdk": "^0.36.0",
     "ink": "^6.5.1",
-    "opentype.js": "^1.3.4",
     "p-limit": "^6.2.0",
     "p-map": "^7.0.4",
     "react": "^19.2.0",
@@ -107,7 +104,7 @@
   "license": "Apache-2.0",
   "author": "varg.ai <hello@varg.ai> (https://varg.ai)",
   "sideEffects": false,
-  "version": "0.4.0-alpha106",
+  "version": "0.4.0-alpha108",
   "exports": {
     ".": "./src/index.ts",
     "./ai": "./src/ai-sdk/index.ts",

package/pricing_fix.txt

@@ -0,0 +1,197 @@
+================================================================================
+PRICING FIX LOG - Verified from Official Provider llms.txt & Pricing Pages
+================================================================================
+Date: 2026-04-22
+Sources: fal.ai/models/*/llms.txt, elevenlabs.io/pricing/api, groq.com/pricing,
+         fireworks.ai/pricing, replicate.com/pricing
+
+================================================================================
+SUMMARY: 13 model definitions updated, 4 confirmed correct
+================================================================================
+
+Models confirmed CORRECT (no changes needed):
+  - reve          ($0.04/image)           src/definitions/models/reve.ts
+  - phota         ($0.09/1K image)        src/definitions/models/phota.ts
+  - phota/edit    ($0.09/1K image)        src/definitions/models/phota.ts
+  - phota/enhance ($0.13/image)           src/definitions/models/phota.ts
+
+================================================================================
+CHANGES APPLIED
+================================================================================
+
+--- FIX #1: nano-banana-pro.ts (CRITICAL - was 3.75x underpriced) ---
+File: src/definitions/models/nano-banana-pro.ts
+Source: https://fal.ai/models/fal-ai/nano-banana-pro/llms.txt
+Old: $0.04 per image flat
+New: $0.15 per image (1K), $0.30 per image (4K)
+     Web search adds $0.015 per request
+Formula: resolution === "4K" ? 0.30 : 0.15, multiplied by numImages
+
+--- FIX #2: recraft-v4.ts (CRITICAL - was 3.1x underpriced) ---
+File: src/definitions/models/recraft-v4.ts
+Source: https://fal.ai/models/fal-ai/recraft/v4/pro/text-to-image/llms.txt
+Old: $0.08 per image flat
+New: $0.25 per image flat
+Formula: flat 0.25
+
+--- FIX #3: kling.ts (was flat, now per-second) ---
+File: src/definitions/models/kling.ts
+Source: https://fal.ai/models/fal-ai/kling-video/o3/pro/text-to-video/llms.txt
+Old: $1.20 flat per generation
+New: $0.112 per second (audio off), $0.14 per second (audio on)
+     5s video = $0.56 (no audio) or $0.70 (audio)
+     Duration range: 3-15 seconds
+Formula: 0.112 * duration
+Note: Using base rate $0.112 (no audio). Audio surcharge not auto-detected.
+      Kling v2.5 Turbo Pro ($0.07/sec) is a different endpoint used in
+      provider code, not captured in this model definition.
+
+--- FIX #4: wan.ts (was flat, now per-second with resolution tiers) ---
+File: src/definitions/models/wan.ts
+Source: https://fal.ai/models/fal-ai/wan-25-preview/image-to-video/llms.txt
+Old: $0.60 flat per generation
+New: $0.05/sec (480p), $0.10/sec (720p), $0.15/sec (1080p)
+     Duration options: 5s or 10s
+Formula: rateMap[resolution] * duration
+         rateMap = { "480p": 0.05, "720p": 0.10, "1080p": 0.15 }
+         Default resolution: "480p"
+
+--- FIX #5: omnihuman.ts (was flat, now per-second) ---
+File: src/definitions/models/omnihuman.ts
+Source: https://fal.ai/models/fal-ai/bytedance/omnihuman/v1.5/llms.txt
+Old: $0.80 flat per generation
+New: $0.16 per second of output video
+     Max audio: 30s (1080p), 60s (720p)
+Formula: 0.16 * duration
+Note: Duration driven by audio length. Output includes billing duration field.
+
+--- FIX #6: veed-fabric.ts (was flat, now per-second with resolution tiers) ---
+File: src/definitions/models/veed-fabric.ts
+Source: https://fal.ai/models/veed/fabric-1.0/llms.txt
+Old: $0.80 flat per generation
+New: $0.08/sec (480p), $0.15/sec (720p)
+Formula: (resolution === "720p" ? 0.15 : 0.08) * duration
+Note: Duration driven by audio length. Resolution is required input.
+
+--- FIX #7: nano-banana-2.ts (was 2x underpriced, now has resolution tiers) ---
+File: src/definitions/models/nano-banana-2.ts
+Source: https://fal.ai/models/fal-ai/nano-banana-2/llms.txt
+Old: $0.04 per image flat
+New: $0.06 (0.5K), $0.08 (1K), $0.12 (2K), $0.16 (4K) per image
+     Web search adds $0.015, high thinking adds $0.002
+Formula: rateMap[resolution] * numImages
+         rateMap = { "0.5K": 0.06, "1K": 0.08, "2K": 0.12, "4K": 0.16 }
+         Default resolution: "1K"
+
+--- FIX #8: elevenlabs.ts (was 1.8x overpriced) ---
+File: src/definitions/models/elevenlabs.ts
+Source: https://elevenlabs.io/pricing/api
+Old: $0.18 per 1,000 characters ($0.00018/char)
+New: $0.10 per 1,000 characters ($0.0001/char) for Multilingual v2/v3
+     Flash/Turbo models: $0.05 per 1,000 characters
+Formula: 0.0001 * characters
+Note: Default model is eleven_multilingual_v2, so v2 rate is used.
+      Flash/Turbo pricing noted in description for reference.
+
+--- FIX #9: ltx-a2v.ts (was flat, now per-megapixel of video data) ---
+File: src/definitions/models/ltx-a2v.ts
+Source: https://fal.ai/models/fal-ai/ltx-2-19b/audio-to-video/llms.txt
+Old: $0.40 flat per generation
+New: $0.0018 per megapixel of generated video data (width x height x frames)
+     Example: 121 frames at 1280x720 = ~112 MP = $0.2016
+Formula: Math.ceil((width * height * numFrames) / 1_000_000) * 0.0018
+         Default: width=1024, height=768, numFrames=121
+
+--- FIX #10: flux.ts (unit changed from per-image to per-megapixel) ---
+File: src/definitions/models/flux.ts
+Source: https://fal.ai/models/fal-ai/flux-pro/v1.1/llms.txt
+Old: $0.05 per image flat
+New: $0.04 per megapixel (rounded up to nearest MP)
+     Standard 1MP image (1024x1024) = $0.04
+Formula: 0.04 * Math.ceil((width * height) / 1_000_000) * numImages
+         Default: width=1024, height=768 -> 1MP -> $0.04
+
+--- FIX #11: whisper.ts / groq (was flat, now per-hour) ---
+File: src/definitions/models/whisper.ts
+Source: https://groq.com/pricing
+Old: $0.03 flat per transcription
+New: $0.111 per hour of audio (minimum 10s billing per request)
+Formula: (Math.max(duration, 10) / 3600) * 0.111
+         Default duration: 60 seconds
+
+--- FIX #12: whisper.ts / fireworks (was flat, now per-minute) ---
+File: src/definitions/models/whisper.ts
+Source: https://fireworks.ai/pricing
+Old: $0.02 flat per transcription
+New: $0.0015 per audio minute (billed per second)
+Formula: (duration / 60) * 0.0015
+         Default duration: 60 seconds
+         $0.0015/min = $0.000025/sec = $0.09/hour
+
+--- FIX #13: qwen-image-2.ts (minor adjustment) ---
+File: src/definitions/models/qwen-image-2.ts
+Source: https://fal.ai/models/fal-ai/qwen-image-2/text-to-image/llms.txt
+Old: $0.04 per image
+New: $0.035 per image (standard endpoint)
+     Pro endpoint: $0.075/image (separate endpoint, not this definition)
+Formula: 0.035 * numImages
+
+================================================================================
+MODELS WITHOUT DEFINITIONS (AI-SDK only, not updated)
+================================================================================
+
+These are registered in src/ai-sdk/providers/ but have no model definition
+files in src/definitions/models/. Verified pricing from llms.txt for reference:
+
+Grok Imagine Video (fal):
+  - text-to-video: $0.05/sec (480p), $0.07/sec (720p)
+  - image-to-video: $0.05/sec (480p) + $0.002/image, $0.07/sec (720p) + $0.002
+  - edit-video:     $0.06/sec (480p), $0.08/sec (720p) (includes input+output)
+
+Sync Lipsync (fal):
+  - $0.70 per minute of video processed
+
+ElevenLabs Music (elevenlabs):
+  - $0.30 per minute of music generated
+
+Replicate Flux models:
+  - flux-1.1-pro: $0.04/output image
+  - flux-dev:     $0.025/output image
+  - flux-schnell: $0.003/output image ($3/1K images)
+
+================================================================================
+UNVERIFIED MODELS (no public pricing available)
+================================================================================
+
+seedance.ts    - PiAPI credit-based pricing, not publicly listed per-second
+heygen.ts      - HeyGen API uses pay-as-you-go credits, no public $/sec rate
+soul.ts        - Higgsfield has no public pricing page
+sonauto.ts     - fal model page returned 404, may be deprecated/moved
+llama.ts       - Groq LLM pricing is negligible, current approx is acceptable
+
+================================================================================
+PRICING PARAMS INTERFACE
+================================================================================
+
+The PricingParams interface (src/core/schema/types.ts) already has all fields
+used in the updated calculate functions:
+  - duration: number       (video/audio seconds)
+  - resolution: string     (e.g. "480p", "720p", "1080p", "1K", "4K")
+  - width: number          (image/video width in pixels)
+  - height: number         (image/video height in pixels)
+  - numFrames: number      (video frame count)
+  - numImages: number      (batch image count)
+  - characters: number     (TTS text length)
+
+No changes were needed to the interface.
+
+================================================================================
+TYPE CHECK
+================================================================================
+
+All 13 model definition files pass TypeScript type checking with zero errors.
+Pre-existing errors in ai-sdk provider files (Uint8Array/BlobPart) are unrelated.
+
+================================================================================
+END OF PRICING FIX LOG
+================================================================================

package/src/ai-sdk/providers/editly/backends/types.ts

@@ -47,11 +47,6 @@ export interface FFmpegRunOptions {
   verbose?: boolean;
   /** Max execution time in seconds (used by cloud backends like Rendi, ignored by local) */
   timeoutSeconds?: number;
-  /** Extra files (e.g. fonts, ASS subtitles) to include alongside inputs.
-   *  When present, cloud backends like Rendi use compressed folder mode
-   *  (input_compressed_folder) to bundle all files together.
-   *  Each entry provides either a `url` to download or raw `data` bytes. */
-  auxiliaryFiles?: { url?: string; data?: Uint8Array; fileName: string }[];
 }
 
 export type FFmpegOutput =

package/src/ai-sdk/providers/editly/rendi/index.ts

@@ -1,4 +1,3 @@
-import { zipSync } from "fflate";
 import sharp from "sharp";
 import { File } from "../../../file";
 import type { StorageProvider } from "../../../storage/types";
@@ -129,11 +128,6 @@ export class RendiBackend implements FFmpegBackend {
   }
 
   async run(options: FFmpegRunOptions): Promise<FFmpegRunResult> {
-    // When auxiliary files (e.g. fonts) are present, use compressed folder mode
-    if (options.auxiliaryFiles && options.auxiliaryFiles.length > 0) {
-      return this.runWithCompressedFolder(options);
-    }
-
     let {
       inputs,
       filterComplex,
@@ -293,204 +287,6 @@ export class RendiBackend implements FFmpegBackend {
     throw new Error("Rendi command timed out");
   }
 
-  /**
-   * Run an FFmpeg command using Rendi's input_compressed_folder mode.
-   *
-   * Used when auxiliary files (e.g. fonts for subtitle rendering) need to be
-   * bundled alongside regular inputs. Creates a ZIP containing all input files
-   * and auxiliary files, uploads it to storage, and submits to Rendi with
-   * `input_compressed_folder` instead of `input_files`.
-   *
-   * Inside the ZIP, all files are at the root level. The ffmpeg command
-   * references files by their bare filenames (not placeholders).
-   */
-  private async runWithCompressedFolder(
-    options: FFmpegRunOptions,
-  ): Promise<FFmpegRunResult> {
-    const {
-      inputs,
-      videoFilter,
-      filterComplex,
-      outputArgs = [],
-      outputPath,
-      verbose,
-      auxiliaryFiles = [],
-    } = options;
-
-    // 1. Resolve all input files to URLs
-    const inputEntries: { fileName: string; url: string }[] = [];
-    for (const input of inputs ?? []) {
-      const path = this.getInputPath(input);
-      const url = await this.resolvePath(path);
-      // Extract filename from URL or path
-      const fileName =
-        url.split("/").pop()?.split("?")[0] ?? `input_${inputEntries.length}`;
-      inputEntries.push({ fileName, url });
-    }
-
-    // 2. Download all files (inputs + auxiliary) into memory
-    const zipContents: Record<string, Uint8Array> = {};
-
-    const downloadTasks = [
-      ...inputEntries.map(async (entry) => {
-        const res = await fetch(entry.url);
-        if (!res.ok)
-          throw new Error(
-            `Failed to download input ${entry.fileName}: ${res.status}`,
-          );
-        zipContents[entry.fileName] = new Uint8Array(await res.arrayBuffer());
-      }),
-      ...auxiliaryFiles.map(async (file) => {
-        if (file.data) {
-          // Inline data — no download needed
-          zipContents[file.fileName] = file.data;
-          return;
-        }
-        if (!file.url) {
-          throw new Error(
-            `Auxiliary file ${file.fileName} has neither url nor data`,
-          );
-        }
-        const res = await fetch(file.url);
-        if (!res.ok)
-          throw new Error(
-            `Failed to download auxiliary file ${file.fileName}: ${res.status}`,
-          );
-        zipContents[file.fileName] = new Uint8Array(await res.arrayBuffer());
-      }),
-    ];
-
-    await Promise.all(downloadTasks);
-
-    if (verbose) {
-      const totalSize = Object.values(zipContents).reduce(
-        (sum, buf) => sum + buf.length,
-        0,
-      );
-      console.log(
-        `[rendi] creating ZIP with ${Object.keys(zipContents).length} files (${(totalSize / 1024 / 1024).toFixed(1)} MB)`,
-      );
-    }
-
-    // 3. Create ZIP
-    const zipData = zipSync(zipContents, { level: 1 }); // fast compression
-
-    // 4. Upload ZIP to storage
-    const zipKey = `internal/rendi-compressed-${Date.now()}.zip`;
-    const zipUrl = await this.storage.upload(
-      zipData,
-      zipKey,
-      "application/zip",
-    );
-
-    if (verbose) {
-      console.log(
-        `[rendi] uploaded ZIP (${(zipData.length / 1024 / 1024).toFixed(1)} MB) -> ${zipUrl}`,
-      );
-    }
-
-    // 5. Build ffmpeg command using bare filenames (not {{in_X}} placeholders)
-    const inputArgs: string[] = [];
-    for (const [i, input] of (inputs ?? []).entries()) {
-      if (typeof input !== "string" && "options" in input && input.options) {
-        inputArgs.push(...input.options);
-      }
-      inputArgs.push("-i", inputEntries[i]!.fileName);
-    }
-
-    const filterArgs: string[] = [];
-    if (filterComplex) {
-      filterArgs.push("-filter_complex", filterComplex);
-    }
-    if (videoFilter) {
-      // For compressed folder mode, the video filter references files by
-      // their bare filenames (already resolved in the working directory)
-      filterArgs.push("-vf", videoFilter);
-    }
-
-    const processedOutputArgs = outputArgs.filter((arg) => arg !== "-y");
-
-    const commandParts = [
-      ...inputArgs,
-      ...filterArgs,
-      ...processedOutputArgs,
-      "{{out_1}}",
-    ];
-    const ffmpegCommand = this.buildCommandString(commandParts);
-    const outputFilename = outputPath?.split("/").pop() ?? "output.mp4";
-
-    if (verbose) {
-      console.log("[rendi] input_compressed_folder:", zipUrl);
-      console.log("[rendi] ffmpeg_command:", ffmpegCommand);
-    }
-
-    // 6. Submit to Rendi with input_compressed_folder
-    const submitResponse = await fetch(`${RENDI_API_BASE}/run-ffmpeg-command`, {
-      method: "POST",
-      headers: {
-        "X-API-KEY": this.apiKey,
-        "Content-Type": "application/json",
-      },
-      body: JSON.stringify({
-        input_compressed_folder: zipUrl,
-        output_files: { out_1: outputFilename },
-        ffmpeg_command: ffmpegCommand,
-        max_command_run_seconds:
-          options.timeoutSeconds ?? this.maxCommandRunSeconds,
-      }),
-    });
-
-    if (!submitResponse.ok) {
-      const errorText = await submitResponse.text();
-      throw new Error(
-        `Rendi submit failed: ${submitResponse.status} - ${errorText}`,
-      );
-    }
-
-    const { command_id } =
-      (await submitResponse.json()) as RendiCommandResponse;
-
-    if (verbose) {
-      console.log("[rendi] command_id:", command_id);
-    }
-
-    // 7. Poll for completion (same as standard run)
-    let attempts = 0;
-    while (attempts < MAX_POLL_ATTEMPTS) {
-      const statusResponse = await fetch(
-        `${RENDI_API_BASE}/commands/${command_id}`,
-        {
-          headers: { "X-API-KEY": this.apiKey },
-        },
-      );
-
-      if (!statusResponse.ok) {
-        throw new Error(`Rendi poll failed: ${statusResponse.status}`);
-      }
-
-      const status = (await statusResponse.json()) as RendiStatusResponse;
-
-      if (status.status === "SUCCESS") {
-        const outputFile = status.output_files?.out_1;
-        if (!outputFile?.storage_url) {
-          throw new Error("Rendi completed but no output URL");
-        }
-        return { output: { type: "url", url: outputFile.storage_url } };
-      }
-
-      if (status.status === "FAILED") {
-        throw new Error(
-          `Rendi command failed: ${status.error_message ?? "unknown error"}`,
-        );
-      }
-
-      await this.sleep(POLL_INTERVAL_MS);
-      attempts++;
-    }
-
-    throw new Error("Rendi command timed out");
-  }
-
   async resolvePath(input: FilePath): Promise<string> {
     if (input instanceof File) {
       return input.upload(this.storage);

package/src/react/elements.ts

@@ -157,7 +157,7 @@ export function Title(props: TitleProps): VargElement<"title"> {
   return createElement(
     "title",
     props as Record<string, unknown>,
-    props.children,
+    props.children ?? props.text,
   );
 }
 
@@ -165,7 +165,7 @@ export function Subtitle(props: SubtitleProps): VargElement<"subtitle"> {
   return createElement(
     "subtitle",
     props as Record<string, unknown>,
-    props.children,
+    props.children ?? props.text,
   );
 }