Description:
This workflow allows you to generate video from text.
You will find a step-by-step guide to using this workflow here: link
My other workflows for WAN: link
Resources you need:
📂Files :
For base version
T2V Model: fp16, fp8
In models/diffusion_models
For GGUF version
T2V Quant Model: Q8, Q5, Q3
In models/diffusion_models
Common files :
CLIP: umt5_xxl_fp8_e4m3fn_scaled.safetensors
in models/clip
VAE: wan_2.1_vae.safetensors
in models/vae
Speed LoRA: lightx2v_T2V_14B_cfg_step_distill_v2_lora_rank64_bf16.safetensors
in models/loras
ANY upscale model:
Realistic : RealESRGAN_x4plus.pth
Anime : RealESRGAN_x4plus_anime_6B.pth
in models/upscale_models
📦Custom Nodes :
Description
What's new? :
Added non-GGUF version and without nightly node,
New interface,
New upscaler,
New model optimisation,
New LoRA loader,
New FLUX version,
New model loader to unload part to RAM.
FAQ
Comments (1)
Can anyone explain why on a 12GB 5070 (what I'm experimenting on) the Kijai FP8 14b T2V is faster in every scenario than the any GGUF regardless of quantization?
Tested the Q3_K_S, Q4_K_S, and Q5_K_S and they're all slower