LTX-2.5 Water Simulation LoRA: add rain or a flood to a clip
LTX-2.5 Water Simulation adds water to a dry reference clip at native 1920x1088. How its prompt works, and the hosted edit route on Sume instead.

LTX-2.5 Water Simulation is an open adapter that adds water to a clip you already have, such as rain, a flood, a river or a splash, while keeping the scene's subject, framing and camera move. On Sume the nearest tool is Gemini Omni Flash 1.1's video edit mode, where a prompt describes the change, but nothing there is specific to water.
The adapter facts are from the Water Simulation card, read on 2026-10-02. Sume's edit facts come from Video Router.
How do you prompt the Water Simulation adapter?
You supply a dry reference clip at 24 fps and a prompt in a fixed shape: the reference shows the scene, the edited version shows the same scene with water added, then ADD WATER and a description. The card's training data is about 572 paired examples of real water events, with the water digitally removed from the reference clips for conditioning.
- Native resolution 1920x1088 or 1088x1920, at 24 fps.
- At least 121 frames, where (frames - 1) % 8 == 0.
- LoRA strength 1.0; values of 1.5 or higher may distort faces and fine detail.
- Stage-1 distilled inference at native resolution; the card says stage-2 upscaling causes identity drift.
What does it not do?
Three limits are stated. There is no mask support, so the effect applies across the whole scene. It was trained on real water, so other liquids generalise inconsistently. And it is not designed for selective application. If you want rain on the street but not on the window, you cannot say so with a mask.
How does a Sume edit compare?
In Video Router, gemini-omni-flash-1.1 accepts a video_url for video_to_video edits. The prompt describes the edit, resolution is optional with a 720p default, and there is no aspect_ratio or duration field. The docs list its video length at 3 to 10 seconds and list no image or reference fields for an edit. Here is the docs' own edit call, with the prompt changed:
curl -X POST https://api.sume.com/v1/video-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: water-edit-001" \
-d '{
"model": "gemini-omni-flash-1.1",
"prompt": "Add heavy rain and wet pavement. Keep everything else the same.",
"video_url": "https://example.com/clip.mp4",
"resolution": "720p",
"mode": "async"
}'Which should you use for a water effect?
Use the adapter when the effect has to sit on a specific 1920x1088 shot with identity preserved and you can run a GPU workflow. Use a hosted edit when you want a quick look and 720p is enough. Sume's docs do not promise that an edit keeps faces intact, so compare the output frame by frame.
Billing for that model is provider list times 1.25 per output second by resolution, as the docs state. Read the edit video modes post for which models accept a source clip. The adapter licence is the LTX-2 community licence.
Sources
Related posts
More in Models
- Luma API camera concepts: 34 named moves vs Sume's prompt-only camera
Luma's API lists 34 named camera concepts you can layer, such as orbit_left and dolly_zoom. Sume has no camera field, so here is how to ask for the same moves.
- Luma Photon API: 4 image refs and a character ref vs Sume references
Luma's photon-1 image API takes up to 4 image references, a style reference and a 4-image character reference, each with a weight. Here is the Sume equivalent.
- Lyria 3.5 is single-turn and varies per call: keep the artifact
Google says Lyria generation is single-turn, not iteratively editable and varies between calls. Why to save every Sume Music artifact you like.
- Lyria 3.5 in another language: prompt in it, then check the song
Google says Lyria 3.5 makes music in other languages when you prompt in that language. On Sume, write the brief and lyrics in it, then check the result.
Written by Sume