Spaces:
Running on Zero
Running on Zero
| title: LTX-2.4 Prompt Enhancer | |
| emoji: ✨ | |
| colorFrom: green | |
| colorTo: blue | |
| sdk: gradio | |
| sdk_version: 6.13.0 | |
| app_file: app.py | |
| pinned: false | |
| hardware: zero-a10g | |
| short_description: Gemma-4 prompt enhancer for LTX-2.4 (called by the demos) | |
| models: | |
| - google/gemma-4-E2B-it | |
| # LTX-2.4 Prompt Enhancer | |
| Standalone Gemma-4 (`google/gemma-4-E2B-it`) prompt enhancer for LTX-2.4. Expands a short user | |
| prompt into a detailed audio-visual caption in the model's training-caption style, exactly | |
| replicating `diffusers`' `LTX2Pipeline.enhance_prompt` (system prompt by mode, greedy decoding | |
| with `no_repeat_ngram_size=3`). | |
| Kept as a separate ZeroGPU Space so the enhancer (and its `torchvision` dependency) stays off the | |
| main 22B video Space's memory budget — the LTX-2.4 video demos call this over `gradio_client`. | |
| Add a first-frame image to enhance for image-to-video. | |