ltx2-videolisted

Generate video from a photo (or two) using self-hosted LTX-2.3 on Modal GPU. THIS is the skill for turning a single photo into a video — prefer it over any video-to-video / image skill whenever the user has a photo and wants motion. Use this whenever the user wants to turn an image into a video, animate a photo, make a reel/clip, do keyframe interpolation between two images, restyle a video (video-to-video / retake), or generate video from a text prompt — even if they don't say the word "video", e.g. "bring this photo to life", "make this move", "animate this", "turn these two shots into a transition". Calls the user's deployed `ltx2-fast-inference` Modal app and saves an .mp4 locally. Triggers: "make a video", "animate this photo", "image to video", "i2v", "keyframe", "interpolate", "video to video", "retake", "restyle this clip", "generate a clip/reel", "follow this pose/edges/depth", "canny/pose/depth control", "match this motion".
patraxo/ltx2-vidgen-skill · ★ 6 · Code & Development · score 69

Install: claude install-skill patraxo/ltx2-vidgen-skill

# ltx2-video — photo → video via self-hosted LTX-2.3 Turns a local image (or two, or a video) into an `.mp4` by calling the user's **deployed** `ltx2-fast-inference` Modal app (LTX-2.3, 22B). Five modes: | Mode | Input | What it does | |---|---|---| | `i2v` (default) | 1 image + prompt | animates the photo into a clip | | `keyframe` | 2 images + prompt | interpolates A → B | | `v2v` | 1 video + prompt | regenerates a time window (retake) | | `t2v` | prompt only | text-to-video, no image | | `control` | control render (+ optional init image) + prompt | IC-LoRA structural control — `union` follows a canny/depth/pose render. Canny auto-derives from a source video via ffmpeg; depth/pose need a pre-rendered control video. | The work is done by `scripts/submit_video.py`, which calls the deployed app's methods remotely via `modal.Cls.from_name` (no repo path needed). ## Setup (one-time) - `pip install modal && modal token new` - The backend must be deployed: `modal app list | grep ltx2-fast-inference`. If absent, deploy it from the `ltx2-fast-inference` repo: `./deploy.sh`. ## Workflow 1. **Resolve + validate the image.** Get the absolute path and confirm it's an image: ```bash realpath "<user-path>" # normalize ~, relative, drag-dropped paths file "<abs-path>" # must contain JPEG / PNG / image data ``` If not found or not an image, report and stop. 2. **Confirm before running** (it costs GPU time). Use **AskUserQuestion**: - hea