LTX 2.5 DEV Image to Video

Description:

Turn a still image into a video clip with LTX 2.5 DEV, Lightricks' 22B open-weights video model.

Upload a photo as the first frame, describe how you want the scene to move, and hit run. The workflow generates video at a lower resolution first, then runs a latent spatial upscaler (2x) and a second refinement pass for sharper detail. Audio generates alongside the video in a single pass. Default output: 5 seconds at 24fps in 16:9 widescreen.

How do you use LTX 2.5 for image-to-video generation?

Upload your image as the first frame, write a prompt describing the motion you want, and run. The workflow handles two-stage generation with latent upscaling built in. Leave prompt enhance on for your first run. Duration, resolution, and frame rate are all adjustable.

Image (first frame)
Your starting frame. Works best with a clear, well-lit photo where the subject and environment are easy to read. Portraits, outdoor scenes, product shots, and concept art all work.

Prompt
Describe what moves and how. Be specific about physical motion: "She turns her head to the left and smiles" gives better results than "a woman moves." Include details about what stays the same (camera position, background, clothing) to keep the scene stable.

Prompt Enhance (default: on)
When this is on, the workflow sends your prompt through Gemma 4 12B to expand it with details the model responds to. Leave it on when you want richer motion from a short description. Turn it off if you wrote a detailed prompt and want the model to follow it closely.

Duration (default: 5 seconds)
How long the output clip runs. LTX 2.5 DEV supports longer clips, but 5 seconds is a good starting point. Keep it short while you dial in the look, then extend once you have settings you like.

Resolution (default: 16:9 Widescreen, 1280x720)
Pick your aspect ratio from the selector. The workflow generates at a smaller internal resolution and upscales 2x in latent space, so the final output is sharper than the raw generation size.

Frame Rate (default: 24fps)
Standard cinematic frame rate. Increase to 30 or higher for smoother motion if you need it for web or social content.

Switch to Text to Video (default: off)
Flip this on to generate video from your prompt alone, with no input image. Useful when you want the model to invent the scene from scratch.

Steps (default: 30)
More steps can improve detail at the cost of longer generation time. 30 is well-optimized. Going lower than 20 may produce softer results.

Seed
Set to random by default. Lock the seed when you want to compare the effect of prompt or setting changes without the randomness shifting the whole output.

What is LTX 2.5 DEV image-to-video good for?

LTX 2.5 DEV is a 22B open-weights model built for speed. It generates synchronized audio and video in a single pass, handles complex motion prompts well, and the latent upscale stage in this workflow keeps the output sharp without a separate upscaling step.

This workflow fits when you need to animate a still frame with specific, directed motion. Product animations where the item rotates or unfolds. Character acting from concept art. Environmental shots where weather or lighting shifts across a scene.

The two-stage pipeline (generate then upscale and refine) means you get cleaner output than a single-pass I2V at the same resolution. The distilled LoRA keeps generation fast while maintaining quality.

LTX 2.5 also generates audio alongside the video. If your scene has ambient sound or action-linked audio (footsteps, wind, impacts), the model attempts to produce matching audio without a separate step.

If you need longer clips or multishot sequences with consistent characters across cuts, LTX 2.5 supports those natively, but you would need a different workflow template for multishot.

FAQ

What resolution does LTX 2.5 DEV output in this workflow?
The workflow generates at a smaller internal size and runs a 2x latent spatial upscaler before the final decode. With the default 16:9 setting, the output lands at 1280x720. The upscale pass adds sharpness that raw single-pass generation at the same size would miss.

Does LTX 2.5 generate audio with the video?
Yes. This workflow loads both a video VAE and an audio VAE. LTX 2.5 generates synchronized audio and video in one pass. The audio matches the visual content, so action scenes get corresponding sound. Quality varies by scene complexity.

Can I use this workflow for text-to-video instead of image-to-video?
Yes. There is a "Switch to Text to Video" toggle. Flip it on and the workflow generates video from your prompt alone, with no reference image. Everything else (prompt enhance, duration, resolution, upscaling) works the same way.

How long does LTX 2.5 DEV take to generate a clip?
Generation time depends on duration and resolution settings. A 5-second clip at 720p with 30 steps takes roughly 30-60 seconds on an H100 GPU. The two-stage pipeline (generate + upscale refine) adds time compared to a single-pass workflow, but the quality improvement is visible.

How do I run LTX 2.5 image-to-video online?
You can run LTX 2.5 image-to-video online through Floyo. No installation, no setup. Open the workflow in your browser, upload your image, and hit run. Free to try.