Boost generation speed by using this lora instead of real CFG. Use BasicGuider or set CFG=1.0
In the examples, top: real cfg only, bottom: split schedule between real cfg and distilled (except the fox, which is middle: distilled only, bottom: no cfg)
It can be used for the full generation schedule, but composition and quality may be worse than real CFG, so for better results, split the schedule to use real CFG without the lora for the first 10-30% of steps, then CFG=1 with the lora for the remainder of steps. Example workflow included in "training data"
Works well with gradient_estimation sampler.
Description
Initial version, trained on videos only with standard Wan video negative prompt, scale=4
FAQ
Comments (18)
is it work with i2v 14B ?
+1 ?
LoRA for T2V of version 1.3B are not capable with I2V/T2V 14B models. Blocks are different
Neat stuff mate!
Hopping for 14b version,
this will change the game for sure.
yeah for image to video too
hope you make the 14b one and btw
if you want to use other loras, which position the lora loader loads?? in the first 10-30% steps or the remainder of the steps.
first 5 step with CFG 4-6 for exemple and 10-15 with CFG 1.0. Best result for me. And the bonus you can cancel if you see bad animations or artefacts
Most useful Lora for Wan 1.3B so far! Huge speed increase with minimum loss of quality.
how to use the model pls can you share the code or repo sorry I am beginner
Can you share the json workflow?
workflow is in the "training data" download
Lora is not loading correctly in comfyui native workflow
lora key not loaded: diffusion_model.blocks.0.cross_attn.k.lora_A.weight
Great work! Could you tell us a little more about the training process?
Better results than the official wan2.1 workflows. lol Man u re awesome
so, I'm using the work flow and I didn't change anything and my outputs are coming out really bad
13B or 1.3B??
13b doesnt exist only 14b lol so its for 1.3b
@gambikules858 1.3b，the bigger one model is 14b（not this one）
Details
Files
wan-1.3b-cfgdistill-video-4.0-00001000_comfy.safetensors
Mirrors
121_wan-13b-cfgdistill-video.safetensors
wan-1.3b-cfgdistill-video-4.0-00001000_comfy.safetensors
wan-1.3b-cfgdistill-video-4.0-00001000_comfy.safetensors
wan-1.3b-cfgdistill-video-4.0-00001000_comfy.safetensors
wan-1.3b-cfgdistill-video-4.0-00001000_comfy.safetensors
wan-1.3b-cfgdistill-video-4.0-00001000_comfy.safetensors
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.