lightx2v

/

Hy1.5-Quantized-Models

like 43

会有的

👍，再是comfyui版本还是不能用，很想用你们新的vae。hunyuan的瓶颈在vae解码，必须分块不说，还慢

vae目前还没适配comfyui，后续会适配。_comfyui结尾的fp8权重还不能用吗？

还是不能用，一样的黑屏，我试的hy15_480p_i2v_distilled_fp8_e4m3_lightx2v_comfyui ,sage2.另外https://github.com/ModelTC/ComfyUI-Lightx2vWrapper 这个项目能适配一下hy1.5吗

对了还有这个项目https://github.com/ModelTC/ComfyUI-LightVAE 😄。这几个项目有冲突，lightvae的包依赖的方式不同会冲突

unet missing: ['time_in.in_layer.scale_weight', 'time_in.out_layer.scale_weight', 'txt_in.input_embedder.scale_weight', 'txt_in.t_embedder.in_layer.scale_weight', 'txt_in.t_embedder.out_layer.scale_weight', 'txt_in.c_embedder.in_layer.scale_weight', 'txt_in.c_embedder.out_layer.scale_weight', 'txt_in.individual_token_refiner.blocks.0.adaLN_modulation.1.scale_weight', 'txt_in.individual_token_refiner.blocks.0.self_attn.qkv.scale_weight', 'txt_in.individual_token_refiner.blocks.0.self_attn.proj.scale_weight', 'txt_in.individual_token_refiner.blocks.0.mlp.0.scale_weight', 'txt_in.individual_token_refiner.blocks.0.mlp.2.scale_weight', 'txt_in.individual_token_refiner.blocks.1.adaLN_modulation.1.scale_weight', 'txt_in.individual_token_refiner.blocks.1.self_attn.qkv.scale_weight', 'txt_in.individual_token_refiner.blocks.1.self_attn.proj.scale_weight', 'txt_in.individual_token_refiner.blocks.1.mlp.0.scale_weight', 'txt_in.individual_token_refiner.blocks.1.mlp.2.scale_weight', 'double_blocks.0.img_attn.qkv.weight', 'double_blocks.0.img_attn.qkv.bias', 'double_blocks.0.img_attn.qkv.scale_weight', 'double_blocks.0.txt_attn.qkv.weight', 'double_blocks.0.txt_attn.qkv.bias', 'double_blocks.0.txt_attn.qkv.scale_weight', 'double_blocks.1.img_attn.qkv.weight', 'double_blocks.1.img_attn.qkv.bias', 'double_blocks.1.img_attn.qkv.scale_weight', 'double_blocks.1.txt_attn.qkv.weight', 'double_blocks.1.txt_attn.qkv.bias', 'double_blocks.1.txt_attn.qkv.scale_weight', 'double_blocks.2.img_attn.qkv.weight', 'double_blocks.2.img_attn.qkv.bias', 'double_blocks.2.img_attn.qkv.scale_weight', 'double_blocks.2.txt_attn.qkv.weight', 'double_blocks.2.txt_attn.qkv.bias', 'double_blocks.2.txt_attn.qkv.scale_weight', 'double_blocks.3.img_attn.qkv.weight', 'double_blocks.3.img_attn.qkv.bias', 'double_blocks.3.img_attn.qkv.scale_weight', 'double_blocks.3.txt_attn.qkv.weight', 'double_blocks.3.txt_attn.qkv.bias', 'double_blocks.3.txt_attn.qkv.scale_weight', 'double_blocks.4.img_attn.qkv.weight', 'double_blocks.4.img_attn.qkv.bias', 'double_blocks.4.img_attn.qkv.scale_weight', 'double_blocks.4.txt_attn.qkv.weight', 'double_blocks.4.txt_attn.qkv.bias', 'double_blocks.4.txt_attn.qkv.scale_weight', 'double_blocks.5.img_attn.qkv.weight', 'double_blocks.5.img_attn.qkv.bias', 'double_blocks.5.img_attn.qkv.scale_weight', 'double_blocks.5.txt_attn.qkv.weight', 'double_blocks.5.txt_attn.qkv.bias', 'double_blocks.5.txt_attn.qkv.scale_weight', 'double_blocks.6.img_attn.qkv.weight', 'double_blocks.6.img_attn.qkv.bias', 'double_blocks.6.img_attn.qkv.scale_weight', 'double_blocks.6.txt_attn.qkv.weight', 'double_blocks.6.txt_attn.qkv.bias', 'double_blocks.6.txt_attn.qkv.scale_weight', 'double_blocks.7.img_attn.qkv.weight', 'double_blocks.7.img_attn.qkv.bias', 'double_blocks.7.img_attn.qkv.scale_weight', 'double_blocks.7.txt_attn.qkv.weight', 'double_blocks.7.txt_attn.qkv.bias', 'double_blocks.7.txt_attn.qkv.scale_weight', 'double_blocks.8.img_attn.qkv.weight', 'double_blocks.8.img_attn.qkv.bias', 'double_blocks.8.img_attn.qkv.scale_weight', 'double_blocks.8.txt_attn.qkv.weight', 'double_blocks.8.txt_attn.qkv.bias', 'double_blocks.8.txt_attn.qkv.scale_weight', 'double_blocks.9.img_attn.qkv.weight', 'double_blocks.9.img_attn.qkv.bias', 'double_blocks.9.img_attn.qkv.scale_weight', 'double_blocks.9.txt_attn.qkv.weight', 'double_blocks.9.txt_attn.qkv.bias', 'double_blocks.9.txt_attn.qkv.scale_weight', 'double_blocks.10.img_attn.qkv.weight', 'double_blocks.10.img_attn.qkv.bias', 'double_blocks.10.img_attn.qkv.scale_weight', 'double_blocks.10.txt_attn.qkv.weight', 'double_blocks.10.txt_attn.qkv.bias', 'double_blocks.10.txt_attn.qkv.scale_weight', 'double_blocks.11.img_attn.qkv.weight', 'double_blocks.11.img_attn.qkv.bias', 'double_blocks.11.img_attn.qkv.scale_weight', 'double_blocks.11.txt_attn.qkv.weight', 'double_blocks.11.txt_attn.qkv.bias', 'double_blocks.11.txt_attn.qkv.scale_weight', 'double_blocks.12.img_attn.qkv.weight', 'double_blocks.12.img_attn.qkv.bias', 'double_blocks.12.img_attn.qkv.scale_weight', 'double_blocks.12.txt_attn.qkv.weight', 'double_blocks.12.txt_attn.qkv.bias', 'double_blocks.12.txt_attn.qkv.scale_weight', 'double_blocks.13.img_attn.qkv.weight', 'double_blocks.13.img_attn.qkv.bias', 'double_blocks.13.img_attn.qkv.scale_weight', 'double_blocks.13.txt_attn.qkv.weight', 'double_blocks.13.txt_attn.qkv.bias', 'double_blocks.13.txt_attn.qkv.scale_weight', 'double_blocks.14.img_attn.qkv.weight', 'double_blocks.14.img_attn.qkv.bias', 'double_blocks.14.img_attn.qkv.scale_weight', 'double_blocks.14.txt_attn.qkv.weight', 'double_blocks.14.txt_attn.qkv.bias', 'double_blocks.14.txt_attn.qkv.scale_weight', 'double_blocks.15.img_attn.qkv.weight', 'double_blocks.15.img_attn.qkv.bias', 'double_blocks.15.img_attn.qkv.scale_weight', 'double_blocks.15.txt_attn.qkv.weight', 'double_blocks.15.txt_attn.qkv.bias', 'double_blocks.15.txt_attn.qkv.scale_weight', 'double_blocks.16.img_attn.qkv.weight', 'double_blocks.16.img_attn.qkv.bias', 'double_blocks.16.img_attn.qkv.scale_weight', 'double_blocks.16.txt_attn.qkv.weight', 'double_blocks.16.txt_attn.qkv.bias', 'double_blocks.16.txt_attn.qkv.scale_weight', 'double_blocks.17.img_attn.qkv.weight', 'double_blocks.17.img_attn.qkv.bias', 'double_blocks.17.img_attn.qkv.scale_weight', 'double_blocks.17.txt_attn.qkv.weight', 'double_blocks.17.txt_attn.qkv.bias', 'double_blocks.17.txt_attn.qkv.scale_weight', 'double_blocks.18.img_attn.qkv.weight', 'double_blocks.18.img_attn.qkv.bias', 'double_blocks.18.img_attn.qkv.scale_weight', 'double_blocks.18.txt_attn.qkv.weight', 'double_blocks.18.txt_attn.qkv.bias', 'double_blocks.18.txt_attn.qkv.scale_weight', 'double_blocks.19.img_attn.qkv.weight', 'double_blocks.19.img_attn.qkv.bias', 'double_blocks.19.img_attn.qkv.scale_weight', 'double_blocks.19.txt_attn.qkv.weight', 'double_blocks.19.txt_attn.qkv.bias', 'double_blocks.19.txt_attn.qkv.scale_weight', 'double_blocks.20.img_attn.qkv.weight', 'double_blocks.20.img_attn.qkv.bias', 'double_blocks.20.img_attn.qkv.scale_weight', 'double_blocks.20.txt_attn.qkv.weight', 'double_blocks.20.txt_attn.qkv.bias', 'double_blocks.20.txt_attn.qkv.scale_weight', 'double_blocks.21.img_attn.qkv.weight', 'double_blocks.21.img_attn.qkv.bias', 'double_blocks.21.img_attn.qkv.scale_weight', 'double_blocks.21.txt_attn.qkv.weight', 'double_blocks.21.txt_attn.qkv.bias', 'double_blocks.21.txt_attn.qkv.scale_weight', 'double_blocks.22.img_attn.qkv.weight', 'double_blocks.22.img_attn.qkv.bias', 'dou

https://huggingface.co/lightx2v/Hy1.5-Distill-Models

kj 提出了一个t2v Lora，放到i2v里面能跑但是出现很明显的变色 https://huggingface.co/Comfy-Org/HunyuanVideo_1.5_repackaged