LTX-2.3 Cinemagraph LoRA: Turn Any Still Image Into a Perfect Looping Video With Selective Motion
Lightricks releases the Cinemagraph IC-LoRA for LTX-2.3, an open-weight adapter that keeps backgrounds frozen while moving only a single element — delivering seamless, natural loops from any still image.
On July 9, 2026, Lightricks released the Cinemagraph IC-LoRA for LTX-2.3 — an open-weight adapter that transforms any still image into a seamless looping video where only one element moves while everything else stays perfectly frozen.
Most video models fight to keep backgrounds still. This one was trained to do exactly that by default.
What Is a Cinemagraph?
A cinemagraph is a still photograph in which a minor and repeated movement occurs — a classic hybrid of photography and video. Common examples include a single strand of hair blowing in the wind, steam rising from a coffee cup, or clouds drifting across an otherwise frozen landscape.
Most video generation models are trained to produce motion everywhere, making it actually difficult to keep backgrounds still. The Cinemagraph LoRA flips this: it was purpose-trained to lock the entire frame except for the region you want to animate.
How It Works
The LoRA is an IC-LoRA (In-Context LoRA) built on top of LTX-2.3's 22B image-to-video model. It uses the trigger word CINEMAGRAPH_MOTION in your prompt to activate the selective-motion behavior.
Recommended workflow:
- Start with a still image (your cinemagraph base)
- Use the
CINEMAGRAPH_MOTIONkeyword in your prompt - Describe what should move and what should stay frozen
- Use CFG scale 2.5–4.5 for best results
Example Prompts
Glasses (beach reflection):
CINEMAGRAPH_MOTION, tripod locked-off static camera, zero camera movement, the man, face, hair, clothing, beach, sky, and background remain completely frozen, only the beach reflection inside his sunglasses moves, reflected time-lapse of ocean waves roll and shimmer in the lenses, everything outside the glasses stays still, seamless natural loop
Jump (man + cow):
CINEMAGRAPH_MOTION, tripod locked-off static camera, zero camera movement, the man and cow remain completely frozen, only the clouds in the sky move in fast time-lapse, clouds drift and roll across the sky, grass and foreground remain still, seamless natural loop
Tears (airplane portrait):
CINEMAGRAPH_MOTION, tripod locked-off static camera, zero camera movement, the woman is frozen, her face, eyes, hair, clothing, seat, and airplane interior remain completely frozen, only the illustrated tear drops move downward along her cheeks like a cutout collage animation, seamless natural loop
Key Details
| Property | Value |
|---|---|
| Model | Lightricks/LTX-2.3-22b-LoRA-Cinemagraph |
| Base Model | LTX-2.3 (22B, Image-to-Video) |
| LoRA Type | IC-LoRA (In-Context LoRA) |
| Trigger Word | CINEMAGRAPH_MOTION |
| Weights | ltx-2.3-22b-lora-cinemagraph-0.9.safetensors (~207 MB) |
| License | LTX-2 Community License |
| Access | Gated (auto-accept on Hugging Face) |
| ComfyUI | Compatible via ComfyUI-LTXVideo custom node |
Download & Usage
The LoRA is available on Hugging Face as a gated model — visit the model page and accept the community license to download:
In ComfyUI, load the LoRA via the standard LoRA loader node after installing the ComfyUI-LTXVideo custom node. Use LTX-2.3 as the base checkpoint and apply the LoRA with weight 0.7–1.0.
Why This Matters
Cinemagraphs have historically required manual masking, frame-by-frame rotoscoping, or dedicated mobile apps (Lightricks itself built Pixaloop/Motionleap years ago for this exact purpose). This LoRA brings professional-grade cinemagraph creation into the open-source AI video pipeline for the first time — no manual masking, no layer compositing, just a prompt.