Live2D LTX 2.3 LoRA
[Live2D]Ltx2.3_rank32.safetensors is a motion LoRA for LTX‑2.3 image-to-video generation. It turns character illustrations into short Live2D-style animations using subtle breathing, blinking, hair movement, clothing movement, and small pose changes. It does not generate or improve character identity—the source image determines appearance.
Usage:
Strength
0: disabled/baselineStrength
1: subtle, safest movementStrength
2: recommended general settingStrength
3: stronger but inconsistent benefitStrength
4: usually unnecessary; may reduce coherence
Recommended workflow: use the same image as the first and final guide for a seamless loop. Describe three to six restrained movements in the video prompt.
Requirements:
LTX‑2.3 compatible base model; tested with the 22B distilled FP8 model
LTX‑2.3 video VAE and text encoders
ComfyUI nodes supporting LTX video, model-only LoRA loading and video output
Qwen3‑VL captioner for automatic motion prompts
Image dimensions divisible by 32, such as 1920×1088 or 1280×736
Approximately 16 GB VRAM for the tested FP8 workflow, with model offloading
Tested settings: 4 steps, CFG 2.0, Euler, SGM-uniform, 49 latent frames, 16 fps, three-second seamless loop.
AB tests:
Description
initial commit
FAQ
Comments (2)
zamn this looks neat, any suggestions on how to prompt this? should i just describe the initial image?
edit: i just tried simply with "a live2d wallpaper of character" and it worked fine!
The best way to prompt it that i came up with is in the workflow attached. If you don't use comfyui, hard to say. It's suppose to work better with verbose captions like:
"A young woman with dark hair, red eyes, and bunny ears lies prone on a concrete surface, her body angled slightly upward as she peers through the scope of a large, white sniper rifle. She wears a blue and white uniform with black gloves and white socks, her expression focused yet calm. A small fox-like creature rests beside her, while a bird perches on her shoulder. Nearby, a yellow thermos with a bird figurine and ammunition boxes are scattered. The background reveals debris and discarded firearms. The camera remains steady, capturing the scene from a low angle that emphasizes her readiness. Soft, natural lighting bathes the scene in muted tones, highlighting the quiet tension of her position."
"A young woman with long, flowing blonde hair and prominent fox-like ears adorned with white floral accents lies serenely on her side in a sunlit bed. Her eyes are half-lidded, gazing upward with a gentle, dreamy expression, her lips slightly parted. She wears a white, collared garment with a dark blue bow at the neck and gold buttons, suggesting a uniform or formal attire. Her arms are loosely draped over her chest, and her body is partially covered by a white blanket. The background reveals a bright window with soft greenery outside, and a small white bird perches on the windowsill. The scene is bathed in warm, diffused light, creating a tranquil, pastel-hued atmosphere with soft shadows and a calm, static composition."
But it's a pain to write as a human. Maybe copypaste this into chatbot wit the prompt image? :
"Treat these observations as a short anime video clip and write one coherent, detailed natural-language video caption 150 words. Describe the subject, appearance, clothing, expression, pose, environment, composition, lighting, color palette, and visual style, but devote at least half of the caption to temporal behavior. Add that character breathes and consequent motion. Confidently describe three to six specific, scene-appropriate motions using active verbs and temporal relationships: breathing, blinking, small head or hand gestures, shifting posture. Describe movement or air, moving light, particles, fire, water, foliage, fabric, smoke, or other environmental animation if applicable."