文章
ltx相关模型下载
1、LTX-2-dev-Q8_0.gguf 21.1 GB
https://hf-mirror.com/QuantStack/LTX-2-GGUF/tree/main/LTX-2-dev
https://modelscope.cn/models/QuantStack/LTX-2-GGUF/tree/master/LTX-2-dev
2、ltx-2-19b-distilled-lora_resized_dynamic_fro09_avg_rank_175_bf16.safetensors 3.58 GB
https://hf-mirror.com/Kijai/LTXV2_comfy/tree/main/loras
https://modelscope.cn/models/Kijai/LTXV2_comfy/tree/master/loras
3、ltx-2-19b-embeddings_connector_distill_bf16.safetensors 2.86 GB
https://huggingface.co/Kijai/LTXV2_comfy/tree/main/text_encoders
https://modelscope.cn/models/Kijai/LTXV2_comfy/tree/master/text_encoders
4、gemma_3_12B_it_fp8_e4m3fn.safetensors 13.2 GB
https://hf-mirror.com/GitMylo/LTX-2-comfy_gemma_fp8_e4m3fn/tree/main
https://modelscope.cn/models/cjc1887415157/LTX-2-comfy_gemma_fp8_e4m3fn/files
5、LTX2_video_vae_bf16.safetensors 2.45 GB
https://huggingface.co/Kijai/LTXV2_comfy/tree/main/VAE
https://modelscope.cn/models/Kijai/LTXV2_comfy/tree/master/VAE
6、LTX2_audio_vae_bf16.safetensors 218 MB
https://huggingface.co/Kijai/LTXV2_comfy/tree/main/VAE
https://modelscope.cn/models/Kijai/LTXV2_comfy/tree/master/VAE
7、ltx-2-spatial-upscaler-x2-1.0.safetensors 996 MB
https://huggingface.co/Lightricks/LTX-2/tree/main
https://modelscope.cn/models/Lightricks/LTX-2/files补充资料
https://modelscope.cn/models/MachineDelusions/LTX-2_Image2Video_Adapter_LoRa/resolve/master/LTX-2-Image2Vid-Adapter.safetensorshttps://github.com/RageCat73/RCWorkflows/tree/main/LTX-2-Audio-Sync-Image2Video-Workflows
https://modelscope.cn/models/Phr00t/LTX2-Rapid-Merges/resolve/master/sfw/ltx2-phr00tmerge-sfw-v5.safetensorshttps://modelscope.cn/models/hf/RuneXX-LTX-2-Workflows/files
✅ LTX-2「锁身份专用 Prompt 模板」（推荐版）
🔹 正向 Prompt（Identity Lock）
single character, same character identity as reference image,
stable facial structure and proportions,
consistent face shape across all frames,
no identity drift, no face morphing,
natural skin texture, realistic facial anatomy,
subtle facial motion, controlled expressions,
minimal facial deformation while speaking,
upper-body framing, camera remains stable,
soft lighting, neutral background,
cinematic realism, high temporal consistency
🔹 时间轴动作（强烈建议一起用）
(0.0–1.2s): character holds a neutral expression, subtle breathing,
head remains mostly still, no speaking
(1.2s+): character starts speaking naturally,
mouth movement synchronized with audio,
expressions remain subtle and controlled
🔹 负向 Prompt（非常关键）
face identity change, facial structure shift,
face morphing, face warping, exaggerated expression,
over-smiling, cartoon face, stylized face,
inconsistent facial features,
asymmetric face, distorted mouth,
uncanny face, plastic skin