Model reference · open weights

ltx-desktop-assets

Available as managed deployment Licence fee Audio FatalErrorVXD · community Text→speech 1 variants 1k dl/mo

ltx-desktop-assets is an open-weight audio or speech model from FatalErrorVXD. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released byFatalErrorVXD
TypeAudio & music
TaskText→speech
Runs withdiffusers
Released2026-05-23
Popularity1k downloads / month
LicenceCommercial licence needed

About

What ltx-desktop-assets is

Asset mirror for the LTX Desktop zero-touch installer. This repository redistributes several open-source model weights and LoRAs under their respective licenses so that the LTX Desktop installer can pull them from a single endpoint without per-user HuggingFace authentication or upstream click-through gates.

Built with Fish Audio. This repository redistributes Fish Audio S2-Pro under the Fish Audio Research License. See fish-audio/LICENSE.md for the verbatim terms and fish-audio/NOTICE for the required attribution.

Read the full model card

⚠️ Non-commercial use only

LTX Desktop is distributed as a non-commercial application. The model weights mirrored here inherit per-component licenses, and at least two of them (Fish Audio S2-Pro, Qwen-Image-Edit-2511-4bit) prohibit commercial use. If you intend to use the contents of this repository commercially:

  1. You may not use the Fish Audio S2-Pro weights for any revenue-generating purpose. Obtain a commercial license from Fish Audio: contact business@fish.audio.
  2. You may not use the Qwen-Image-Edit-2511-4bit weights commercially (CC-BY-NC-SA-4.0). Use the official Apache 2.0 Qwen-Image-Edit weights directly from Qwen instead.
  3. ACE-Step (MIT) and the LoRA collection are commercial-safe; individual LoRA licenses are documented per file.

By downloading from this repository you accept the per-component license terms below.

What's in here

PathComponentUpstreamLicenseSize
fish-audio/s2-pro/Fish Audio S2-Pro TTSfishaudio/s2-proFish Audio Research License (non-commercial)~11 GB
qwen-edit/model/Qwen-Image-Edit-2511 (4-bit NF4)ovedrive/Qwen-Image-Edit-2511-4bitCC-BY-NC-SA-4.0~15 GB
loras/ltx/LTX-2.3 workflow LoRAs (Foley-V2A, Cinemagraph — Lightricks)Lightricks / per-LoRAPer-LoRA (documented in loras/README.md)variable
ltx-2.3-22b-ic-lora-ingredients-0.9.safetensors (repo root)LTX-2.3 IC-LoRA — Ingredients (multi-subject reference-sheet video)Lightricks/LTX-2.3-22b-IC-LoRA-Ingredients (gated upstream)LTX-2 Community License (ltx-ic-lora/LICENSE)~1.3 GB
manifest.jsonInstaller manifest (versions, sizes, hashes, license refs)This repo's own< 10 KB

What's NOT in here (and why)

  • LTX-2.3 main weights + most IC-LoRAs + Gemma-3 text encoder + Z-Image-Turbo + Lightricks' own LoRAs — already auto-downloaded by the LTX Desktop installer from Lightricks' own (non-gated) HF repos. No friction to remove. Exception: the Ingredients IC-LoRA is mirrored here (repo root) because its upstream repo is gated; mirroring it keeps the in-app on-demand download click-through-free. Its verbatim license + attribution are in ltx-ic-lora/.
  • Whisper.cpp, Cocktail-Fork MRX, DPT-Hybrid MiDaS, YOLOX, DW-Pose — all MIT/permissive, anonymously downloadable from upstream HF/GitHub. No value in mirroring.
  • ACE-Step models — MIT-licensed, anonymously downloadable from ACE-Step/* on HF. Pinned to a specific commit SHA in manifest.json instead of mirroring.
  • Qwen3.6-VL — distributed via Ollama registry, not HuggingFace. Pulled by the LTX Desktop installer via ollama pull.

The manifest.json file

Single source of truth for the LTX Desktop installer. Lists every component it needs to acquire (whether from this repo or upstream), with version pins, expected sizes, file paths, and license metadata. The first-launch consent screen reads the license refs from here so the user sees one consolidated license-acceptance flow.

How redistribution works under these licenses

  • Fish Audio Research License explicitly permits redistribution for research and non-commercial purposes, subject to mandatory attribution. The verbatim license is at fish-audio/LICENSE.md, the required NOTICE string is at fish-audio/NOTICE, and the required "Built with Fish Audio" attribution appears at the top of this README and in the LTX Desktop UI.
  • CC-BY-NC-SA-4.0 (ovedrive Qwen-Image-Edit-2511-4bit) permits redistribution with attribution, non-commercial use only, and share-alike on derivatives. The license is at qwen-edit/LICENSE and attribution to the original quantizer is at qwen-edit/ATTRIBUTION.md.
  • Per-LoRA licenses (CivitAI sources) are tracked individually in loras/README.md with link-back to each LoRA's original page.
  • LTX-2 Community License (Lightricks Ingredients IC-LoRA) permits redistribution; only Entities with annual revenues ≥ USD $10M need a separate commercial license, so non-commercial / sub-threshold use is covered. The verbatim license is at ltx-ic-lora/LICENSE and attribution at ltx-ic-lora/NOTICE.md.

Status

🚧 Repository skeleton — model weights upload pending. This README and the directory scaffolding are in place; per-component README/LICENSE files and the actual weights are being added in stages. Watch this space.


This mirror is maintained as part of the LTX Desktop project for the express purpose of enabling a zero-touch installer experience for non-commercial users. It is not affiliated with or endorsed by Fish Audio, Alibaba/Qwen, ACE-Step, Lightricks, or any other upstream project named above. All copyrights remain with their respective owners.

From the published model card. Full card on the HuggingFace links in the sidebar.

How it works

How audio & music work

Audio or textinputAudio modelrecognise / synthesiseText or audiooutputSpeech-to-text turns audio into text; text-to-speech and music models turn text into audio.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys ltx-desktop-assets for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (ltx-desktop-assets below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -F model="ltx-desktop-assets" -F file=@audio.mp3

Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms