LTX-2 STUDIO

Every frame tells a story — we direct it.

Cinematic Image-to-Video Studio · Identity-Locked Faces · Director Timelines ZeroGPU-native · Telegram-controlled · Enterprise-audited

⚡ TL;DR — Upload a face → get a cinematic animated clip with that exact identity, RIFE-smoothed motion, Real-ESRGAN-sharp detail, and an 11-language voiceover — from the web UI or a single Telegram caption.

✨ Feature Constellation

🧭 App Navigation Tour

Space → Generate	Space → Gallery & Presets

Telegram Menu	Telegram Control Panel (41 buttons)

(SVG previews above are vector wireframes of the live neon-dark theme — swap in PNG captures any time under assets/screens/.)

🗺 Blueprint — System Architecture

flowchart LR
    subgraph Clients
        U["👤 Web UI<br/>gr.Blocks"]
        B["🤖 Telegram Bot<br/>Termux"]
    end
    subgraph "HF Space (ZeroGPU a10g)"
        G["Gradio API<br/>18-slot contract"]
        P["LTX-2 Distilled fp8<br/>+ ICLoraPipeline"]
        POST["Post chain<br/>RIFE → ESRGAN → TTS mux"]
        V["Vault writer<br/>🔒 locked jsonl"]
    end
    subgraph "HF Hub (open source)"
        M1["Lightricks/LTX-2 19B"]
        M2["Camera LoRAs ×7"]
        M3["IC-LoRA Skill Packs"]
        M4["RIFE + ESRGAN + MMS"]
        DS[("private Dataset<br/>ltx2-space-vault")]
    end
    U --> G --> P --> POST --> V --> DS
    B <-->|"keep-alive REST"| G
    B --> DS
    M1 & M2 & M3 & M4 -.boot-prefetch/on-demand.-> P

 ┌────────┐   18-slot JSON   ┌───────────────────────────────┐
 │ Web UI │ ───────────────▶ │ Gradio /generate_video         │
 └────────┘                  │  ├─ LTX-2 fp8 distilled (S1→S2)│
 ┌────────┐    keep-alive    │  ├─ IC-guide branch (Detailer) │
 │Tele Bot│ ◀──────────────▶ │  ├─ RIFE 2× ─ ESRGAN 2×       │
 └────────┘   session SSE     │  └─ MMS-TTS mux (replace/mix) │
      │                       └──────────────┬────────────────┘
      ▼                                      ▼
 private Dataset VAULT ◀──────────── metadata.jsonl (+ mp4 shards)

🏗 Repository Diagram

flowchart TD
    ROOT["📦 hzwer-lineage Space repo (this)"] --> APP["app.py — Studio UI + pipeline glue"]
    ROOT --> PKG["packages/ ltx-core + ltx-pipelines (official)"]
    ROOT --> RIF["rife_ifnet.py (verbatim v4.25)"]
    ROOT --> ESR["esrgan_rrdbnet.py (RRDBNet x2)"]
    ROOT --> ASSETS["assets/ banner + nav screens"]
    ROOT --> REQ["requirements.txt · packages.txt"]

    BOT["📱 ltx2_telegram_bot.py (Termux host)<br/>same 18-slot API contract"] -.REST/SSE.-> ROOT
    BOT --> ADM[".ltx2_admins.json · .ltx2_bot_settings.json"]

    VAULT[("🗄 ltx2-space-vault (private dataset)")] <-."vault API".-> APP
    VAULT <-."gallery/presets".-> BOT

🚀 Quick Start

Space (zero-setup): open the Space → Generate tab → drop a portrait → describe motion → Generate. Pro knobs live in ✨ Post-Processing and Advanced accordions; every control is also a REST slot (18-param contract).

Telegram remote:

termux-wake-lock
pip install requests pillow huggingface_hub
python ~/ltx2_telegram_bot.py          # first /start claims 👑 ownership
/menu                                   # full button console
/photo + caption "slow push-in #up #rife #pack_detail #anchor_balanced"
/vonarr नमस्ते दुनिया                    # attach narration (11 languages)

🛠 Development Roadmap

gantt
    title Delivery timeline (completed ✔ / planned ▢)
    dateFormat YYYY-MM-DD
    section Core
    Distilled fix + identity-lock      :done, 2026-08-25, 1d
    ZeroGPU VRAM compliance            :done, 2026-08-25, 1d
    SDPA attention routing             :done, 2026-08-25, 1d
    section Growth
    Post-processing suite              :done, 2026-08-26, 1d
    IC-guide Detailer wiring           :done, 2026-08-26, 1d
    Enterprise UI + audit              :done, 2026-08-26, 1d
    section Next
    Pose/Union preprocessors           :2026-08-27, 3d
    LoRA training runbook (cloud)      :2026-08-30, 5d
    Full-model CFG quality mode        :2026-09-05, 7d
    Moderation + public hardening      :2026-09-10, 5d

✔ Shipped: identity-lock engine · fp8 ZeroGPU fit · SDPA routing · IC-guide Detailer · RIFE/ESRGAN/TTS suite · Vault+Presets · 41-button Telegram parity · enterprise audit fixes · keep-alive perf (3.3×)

▢ Next queue: control-map preprocessors (depth/pose) → real Pose/Union guides · cloud-GPU LoRA training runbook (LTX Trainer) · full-model CFG path · moderation gate · i18n strings

📸 Proof of Work

[boot] StudioTheme constructed
RIFE weights ready + verified: …/MonsterMMORPG--Practical-RIFE/…/flownet.pkl
ESRGAN x2 weights ready: …/ai-forever--Real-ESRGAN/…/RealESRGAN_x2.pth
IC-guide mode: building reference video…
Enhanced prompt (identity-locked): …
event: complete
[vault] saved runs/2026-08-26/88acb57929.mp4
[post] voiceover (Hindi, Mix over native audio) ok
vault latest pack: "Detailer [IC-guide]" · post: 'rife,upscale,vo:Hindi'

🔐 Security & Licensing

• Secrets read from env first (TG_BOT_TOKEN / HF_TOKEN) — rotate the historical tokens shared during development
• Owner-claim auth on the bot; dangerous ops (/reboot, /reboot?factory, logs) admin-only; flood caps (1 queued + 1 running/user, global 8, 60 s cooldown)
• Model licenses: base Apache-2.0 (LTX-2 open weights) · camera LoRAs per Lightricks terms · MMS-TTS = CC-BY-NC (non-commercial) · MusicGen alternative available if commercial use is needed

📚 References

1. LTX-2.5 official workflow guide — multiscale S1/S2 anchor semantics, CFG rules, frame laws
2. Practical-RIFE (hzwer) + weights hzwer/RIFE, mirror MonsterMMORPG/Practical-RIFE
3. ai-forever/Real-ESRGAN — RRDBNet ×2 weights
4. facebook/mms-tts — VITS multilingual TTS
5. Lightricks IC-LoRAs — Pose / Union / Detailer
6. smart_vision_comfyui_ltx25_director — camera-intent vocabulary inspiration (MIT)
7. Skills DB used for UX audit: ui-ux-pro-max · frontend-design · huggingface-gradio/spaces

🙌 Credits

Built on Lightricks LTX-2 open weights · RIFE by Hzwer · Real-ESRGAN by Xintao Wang · MMS by Meta · ComfyUI director concepts by Smart Vision · Telegram + Gradio integration, identity-lock system, Director timelines, vault layer and audits crafted for @tabrezahmed51.