---
title: JoyAI-Echo x LTX-2.3 Multi-Shot Video
emoji: 🎬
colorFrom: yellow
colorTo: red
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
short_description: Multi-shot narrated video with cross-shot memory
python_version: "3.12"
startup_duration_timeout: 1h
pinned: false
license: other
models:
- google/gemma-3-12b-it
- joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical
- joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical-gguf
- joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical-int8
- joeygambino/joyai-echo-multishot-workflow
---

# JoyAI-Echo × LTX-2.3 — Multi-Shot Narrated Video

Joint **audio-video** generation with cross-shot character consistency, using the
community *surgical merge* checkpoint
[`joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical`](https://huggingface.co/joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical).

The merge keeps:

- **JoyAI-Echo's** video / conditioning branch — its slot-paired cross-modal
  **memory bank**, which holds a character's face/wardrobe/voice identity across
  shots.
- **LTX-2.3-distilled-1.1's** audio branch — natural voice and lip-sync.

This Space runs the **official JoyAI-Echo inference graph** (`ltx-core` /
`ltx-pipelines` / `ltx-distillation`) with the merged **fp8** checkpoint (~23.4 GB)
dropped in as the single `model_file`, so it fits the ZeroGPU budget. The
pipeline's own 3-phase module hot-swap keeps peak VRAM manageable.

Write your story as one or more **shots**, separated by a line containing only
`---`. Reuse the same character description (e.g. `ID_A is …`) across shots so the
memory bank preserves identity and voice.

- Text encoder: **Gemma-3-12B-it** (gated; needs an `HF_TOKEN` secret with access).
- **Non-commercial** use only (LTX-2 Community License; the JoyAI-Echo component is
  research/non-commercial). Generated content is machine-generated and should be
  disclosed as such.

## Model variants

| build | for | where |
| --- | --- | --- |
| bf16 / fp8 (this demo runs the fp8) | reference / 40-50-series | [base repo](https://huggingface.co/joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical) |
| Q8_0 / Q5_0 GGUF | JoyAI-Echo multishot ComfyUI pack, highest quant fidelity | [GGUF repo](https://huggingface.co/joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical-gguf) |
| INT8 ConvRot | stock ComfyUI >= 0.27, any RTX 20-series+, fastest | [INT8 repo](https://huggingface.co/joeygambino/joyai-echo-ltx23-echoVid-ltxAud-surgical-int8) |

## Disclosure

Outputs are AI-generated video with AI-generated audio. The LTX-2 Community
License requires machine-generated content to be disclosed as such wherever it
is shared; the JoyAI-Echo component is research / non-commercial.