Tomato AI: One Sentence, One Video

From text to video, from image to short film, from generation to editing — full workflow, zero barriers, no watermark. This article takes you behind the scenes of Tomato AI (cctocv.com) and its product design and engineering.

---

Why Do We Need Another AI Video Tool?

In 2025, the AI video generation space is crowded: Runway, Pika, Kling, JiMeng... every platform competes on models, parameters, and duration. But creators' pain points remain unchanged:

• Watermarks — Free versions are plastered with logos; commercial use costs extra
• Queues — Popular models make you wait 30+ minutes
• Fragmentation — After generation, you must download, import into Premiere/CapCut, then edit
• High barrier — English interfaces, complex parameters, steep learning curves
• Opaque pricing — You buy a membership only to find pay-per-use fees on top

Tomato AI's answer: combine "generation + editing" into one product, use transparent credit-based billing, aggregate multiple models to eliminate queues, and promise zero watermarks to give the work back to you.

---

Core Capabilities: What Can One Sentence Do?

Five Generation Modes

Open cctocv.com and the floating generator provides a complete creative entry point:

Mode	Description	Use Case
Text to Video (T2V)	Input text description, generate video directly	Creative ideas from scratch
Image to Video (I2V)	Upload image + text, bring static images to life	Product demos, photo animation
Reference Video (Ref2V)	Upload reference material, style-transfer to generate new video	Consistent style series
Image to Image (I2I)	AI image generation, supports 1:1 / 3:2 / 2:3	Covers, graphics, assets
Built-in Video Editor	Timeline, layers, effects, audio — all browser-based	Post-production and compositing

10+ Top-Tier Models, One Entry Point

Tomato AI doesn't build its own models — it's a model aggregation platform that unifies access to ByteDance's JiMeng, Seedance 2.0, Google Veo 3.1, OpenAI Sora 2, Kuaishou Kling 3, and other top-tier models:

• JiMeng 3.0 1080P — ByteDance official model, 10 credits/second, supports first/tail frame control
• Seedance 2.0 (Pippit) — Native audio-video sync, 20 credits/second, supports multi-image reference
• Veo 3.1 / Sora 2 / Kling 3 — International top-tier models on demand

Creators don't need to switch between multiple platforms or compare prices — one credit pool works across all models.

@mention Smart Reference System

This is a standout detail of Tomato AI. In the prompt input box, you can reference uploaded images using @imagename:

A mechanical butterfly flutters through a flower field, its wings formed from @reference.png texture, the camera follows the butterfly through @garden.jpg flower sea

The system auto-pops a candidate dropdown, supports keyboard arrow navigation, and automatically cleans up prompt references when an image is removed. This interaction makes multi-image referencing as natural as mentioning someone in chat.

---

Five Killer Features: Model Capabilities Showcase

The homepage's "Core Features" module showcases the platform's capabilities with real generated examples:

Native Audio-Video Sync (Native Audio-Video Joint Generation)

Models like Seedance 2.0 produce audio simultaneously with video — dialogue, ambient sound, music all in one step, eliminating the disconnect of post-dubbing. A female soldier biting a burger in a war scene: chewing sounds and environmental explosions are generated in sync.

Director-Level Control

Not a random generator, but a creative tool. Camera movement control — aerial dives, tracking shots, orbit close-ups — can all be precisely controlled via prompts. A lone car on a mountain road: the camera goes from aerial overhead to ground-level tracking, all driven by text.

Multi-Shot Storyboarding (Single Prompt)

Generate a complete storyboarded short film from one sentence. A cyberpunk girl fighting a cyborg opening scene — multiple shot transitions, coherent action, clear narrative, all from a single description input.

Realistic Physics Simulation

Time stops, collision dynamics, fluid physics — the model's understanding of real-world physics is astonishing. A galloping horse suddenly "freezes in time," then resumes sprinting after a delay — this effect previously required expensive post-production VFX.

Lightning-Fast Generation

Rendering pipeline optimization delivers extreme speed. An FPV drone flying through a complex Japanese castle scene: from submission to finished video in just minutes.

---

Technical Architecture: Powering a Full-Workflow Creative Platform

Tech Stack

Frontend: Next.js 16 (App Router) + React 19 + Tailwind CSS v4
State: Zustand + React Context
Auth: Firebase Auth (Google OAuth + email/password) → Backend JWT
Payments: PayPal SDK (one-time purchase + monthly subscription)
i18n: i18next (zh/en/ja/ar)
Editor: Custom GPU rendering engine + IndexedDB local persistence
Backend: Express + MySQL (sibling repo videoAaiB)

Key Engineering Decisions

1. Model Abstraction Layer — Unified Interface, Decoupled Differences

Different models have vastly different parameters: JiMeng supports first/tail frame control, Seedance supports multi-image reference and audio-video sync. The code uses MODEL_IMAGE_CAPACITY and MODEL_COSTS_PER_SEC config tables to consolidate differences at the data layer:

const MODEL_IMAGE_CAPACITY = {
  'jimeng_t2v_v30_1080p': 'single',    // single image reference
  'pippit_iv2v_v20_cvtob_with_vinput': 'multi',  // multi-image reference
};

const MODEL_COSTS_PER_SEC = {
  'jimeng_t2v_v30_1080p': 10,   // 10 credits/second
  'pippit_iv2v_v20_cvtob_with_vinput': 20,  // 20 credits/second
};

The UI auto-switches settings panels based on the model — JiMeng shows ratio/duration/resolution/sound/frame control, Seedance shows ratio/duration. Adding a new model requires just one config line.

2. In-Browser Video Editor — opencut Port

This is the platform's heaviest technical investment. Tomato AI embeds a complete video editor (path /editor), derived from the open-source opencut project, deeply customized and ported to Next.js:

• GPU Rendering Engine — WebGL acceleration with automatic fallback on failure
• Timeline Editing — Multi-track, multi-layer, frame-precise
• Effects System — Blur and other real-time filters, based on WGSL shaders
• Local Persistence — IndexedDB stores project files, no server needed
• URL Sync — syncEditorUrl directly manipulates window.history, bypassing the router

The editor manager architecture is clear: playback, timeline, scenes, project, media, renderer, command (undo/redo), save, audio, selection, clipboard, diagnostics — each manager is an independent class coordinating with React's useEditor(selector) hook via subscribe(onChange).

3. Performance Optimization — Video Lazy Loading

The homepage has many showcase videos; loading them all would kill performance. Tomato AI uses IntersectionObserver for three-tier lazy loading:

• Inspiration gallery videos: start loading within 200px of viewport
• Feature showcase videos: preload within 400px, play on hover
• Background video: preload metadata only, poster image fills first

4. Payment System — Dual-Mode Billing

PayPal integration supports two modes:

• One-time purchase — Starter ($20) / Creator ($40) / Studio ($60), credits never expire
• Monthly subscription — Free / Lite / Pro / Premium, PayPal Plan ID auto-distinguishes sandbox and production

Subscription plan IDs are read from environment variables — switching environments requires zero code changes.

5. Internationalization — Four-Language Coverage

zh (Chinese) / en (English) / ja (Japanese) / ar (Arabic), with non-English languages lazy-loaded to reduce initial payload. 27 editor components integrate i18n; Dashboard and Editor have built-in language toggle buttons when there's no global navbar.

---

Product Experience: The Complete Journey from Landing to Finished Film

Homepage — Reducing Decision Cost

Open cctocv.com and you see:

• Video Background Hero — Full-screen AI-generated video as background, instantly establishing "this is what AI can do"
• Floating Generator — Shrinks to a bottom bar when scrolling, expandable anytime, never interrupts browsing
• Five Feature Showcases — Each feature paired with a real generated video, plays on hover, "Use Prompt" jumps directly to generation
• Inspiration Gallery — 7 curated examples, click "Use Prompt" to load into the generator with one click
• Transparent Pricing — One-time vs monthly comparison, annual savings displayed upfront
• FAQ — Common questions answered upfront

Dashboard — Creative Workspace

After login, enter /dashboard. The left navigation includes:

• Home (Explore) — Discovery and recommendations
• Text to Video / Image to Video / Reference Video — Three video generation modes
• Image to Image — AI image generation
• Video Editor — Enter the built-in editor
• History — All generation records
• Settings — Account management

Each generation mode shares the same GeneratorCard component, switching behavior via generationMode to ensure interaction consistency.

Editor — Seamless Transition from Generation to Editing

This is the core differentiator against all competitors. In Tomato AI, you don't need to:

• Download the generated video
• Open another editing software
• Re-import materials

After generation, enter the editor directly in the browser — timeline dragging, transition adding, audio adjusting, final export — the full workflow is a closed loop.

---

Business Model: Credit-Based, Transparent Billing

Credit System

• JiMeng 3.0: 10 credits/second of video
• Seedance 2.0: 20 credits/second of video (includes native audio-video sync)
• AI Image: 10 credits/image

Credits are universal across all models, with no model lock-in. A 5-second JiMeng video = 50 credits, a 15-second Seedance video = 300 credits.

Pricing Plans

Plan	Price	Credits	For
Free	$0	Free on signup	Try it out
Starter (one-time)	$20	1000 credits/180 days	Occasional creation
Creator (one-time)	$40	2000 credits/365 days	Regular creators
Studio (one-time)	$60	3000 credits/365 days	High-volume output
Lite (monthly)	From $0.08/video	100 credits/month	Light subscription
Pro (monthly)	From $0.06/video	330 credits/month	Professional creators
Premium (monthly)	From $0.05/video	800 credits/month	Studio-level

All plans include: full model access, no watermark, commercial license.

---

Final Thoughts: Democratizing AI Video Creation

Tomato AI's vision isn't to be another model company, but to be the bridge between creators and models:

• Model providers focus on pushing quality to the limit
• Tomato AI focuses on pushing the experience to the limit — multi-model aggregation, transparent billing, built-in editing, zero watermark

When generation quality is no longer a barrier (all top-tier models are rapidly improving), workflow efficiency becomes the true differentiator. The shorter the distance from "one sentence" to "a short film," the higher the creator's value.

One sentence to a film, delivered in minutes, commercially usable without watermark — that's Tomato AI.

---

Experience it at: cctocv.com

Tech Stack: Next.js 16 · React 19 · Tailwind v4 · Firebase · PayPal · IndexedDB · WebGL

Supported Languages: 中文 / English / 日本語 / العربية