Generates WAN 2.1 videos in a fraction of time.
720p and 480p Version
Recommended settings:
Sampler/Scheduler: Euler/Simple
Steps: 4
CFG: 1
Sigma-Shift 5
Original Model from Lightx2v converted to FP8 quantisation.
☠️ Do not use any extra speed-up tricks or LoRAs or it may mess up your generations ... 🤬
⚠️ Hint: Most of the time the model is taking you by word. If you write "white" it is white. "Translucent" is translucent... like for the fluids. 💦 Now you know! 🫵 translucent whitish 🤫
⬇️⬇️⬇️⬇️⬇️⬇️⬇️⬇️⬇️⬇️⬇️
Recommended specs:
8 GB VRAM, 32 GB RAM
Sample times: <2 minutes for 81 frames, 4 steps on RTX 4070 Ti Super.
Compatible with 14B LoRAs.
I normally use 0-2 LoRAs, strength at 0.4-1 depending on how much the effect should be. 0.7-0.9 works best most of the time, not overwriting the style of an image.
At multiple LoRAs is seems best to tune the strength a bit down to 0.3-0.6.
Basic workflow example:
Here: https://civarchive.com/models/1811161?modelVersionId=2049602
My favourite UI:
SwarmUI https://github.com/mcmonkeyprojects/SwarmUI
Testing (my specs):
I can go wild on setting with this full checkpoint, even with added LoRAs:
121 frames possible: ~ 3 minutes
121 frames on 24 fps possible (more motion): ~ 3 minutes
128 frames on 24 fps possible (more motion and extended): ~ 3.5 minutes
Dependencies:
YOU are responsible for outputs as always! If you make ToS violating content and I get aware I WILL report this.
Disclaimer
This models are shared without warranties and with the condition that it is used in a lawful and responsible way. I do not support or take responsibility for illegal, harmful, or harassing uses. By downloading or using it, you accept that you are solely responsible for how it is used.
Description
I2V-Model
Wan2.1-I2V-14B-480P-StepDistill-CfgDistill-Lightx2v-fp8
FAQ
Comments (32)
is there a workflow available to help get started?
It is a safetensor model, you can use any UI you want with the normal Wan dependencies, nothing really to consider. Except the ones I listed. LCM, CFG 1, Shift 5. You can even use most 14B LoRAs out there with 0.5-1.
I use swarmui, so I do not use any comfyui workflows directly.
darksidewalker where do i change the shift ? is it somewhere in the ksampler ?
nvm i found. Is is the ModelSampling
I created a very basic workflow example here: https://civitai.com/models/1811161?modelVersionId=2049602
Unfortunately, "Incompatible" when loading into my UI. I'm on Mac M1 Silicon
Most AI sofware are not doing well on Mac. This is just a normal fp8 quant of a Wan 2.1 i2v model ... If you have a "incompatible" problem you may reach out to Lighx2v, they made the original high quant model.
trying a basic i2v, the one built into comfyui, i'm getting a black screen output. EDIT: so I guess I needed to restart comfyui? Because it just started working... okay....
Hi bro, what did you do? I also have black screen output but restarting comfy do not help ;/.
same here. don't matter what, i get only black outputs.
GrzesMis I restarted comfyui and it just started working. Now I have the same issue with wan2.2 and nothing seems to work. Oddly it worked at first and now it doesn't.
Yeah same, getting black screen outputs as well, it worked for some days then suddenly its all just black vids
The model does not change, must be your settings
Sexybeast10 idk why, but when it happens to me. I have noticed that denoise somehow got set to 0. Changing it back to 1 solved my issue.
delta45424155 Ehm there is no denoise setting anywhere on mine o_0 Opening comfy in SwarmUi sometimes fixes it for me, but it just random give me black screen outputs without having made any changes.
clip AND text encoder?
ain't those the same?
are these specific to using Lightx2v or are they the same as normal WAN?
Same, just other quant. In my understanding you could use any other.
Still confused. Do I need both? Are they both checkpoints or is the Clip model a Text Encoder?
dcham2310 The clip is the text encoder. There is no way to just add it to civitai as extra file.
Yep, I downloaded the suggested workflow, corrected the model paths and swapped the Clip to the one you have uploaded here and I get a decent looking 512x512 video clip in under 2 minutes. My 16GB VRAM was at 95% so I suspect I can't go much above that resolution but I can always upscale.
GGUF Q3 and Q2 soon? Thank you!
It's 2025, shouldn't these all be safetensors files? Aren't there well known security risks with pickletensor files (.pth)?
Edit: I'm still trying to figure out the quirks, but this seems like a huge upgrade from FusionX. Only anime previews are shown here, it's not limited in that way at all.
Currently using clip_vision_h and umt5_xxl_fp16. LCM SGM_Uniform or Simple. 4-8 steps. RIFE VFI with a 3 multiplier.
Steps are steps, it doesn't run noticeably faster, but there's a lot more motion. You can have a lower FPS and/or use frame interpolation and get longer videos, or you could do all of that and use a shorter length value and cut down generation time.
I'm getting much more motion and better prompt adherence. I'm having some issues with physics, I think that's just me figuring it out.
I don't use them in comfyui. I use wan21UMT5XxIFP32_fp32 and clip_vision_h.
delta45424155 Thanks! So, I have umt5_xxl_fp8 and 16. Trying to search to umt5_xxl_fp32 orwan21UMT5XxIFP32_fp32 does provide some results, but nothing named clearly named like that. Is that from callgg / upt5-xxl-encoder-fp32 / model.safetensors? If so, what kind of vram damage does it do at 22gb?
Lightx2v released them that way. You can use a other clip and T5. They just have the same quant as the checkpoint.
rando2048 I don't remember where I got it, but am thinking of trying smaller sized ones. it works fine with 16gigs of vram on a 5080. I am limited to short length videos, but you can get around that by saving the last frame and starting another render with it.
darksidewalker The ones you recommend don't work in comfyui. Even if I try to just use the text encoder one; it throws an error. mat1 and mat2. The other file with the .ph. gives me an error that it doesn't contain a valid vision model
rando2048 I just tested umt5-xxl-encoder-q5_k_m.gguf and it works fine. I only tested it on the video I shared.
I'm uploading a working scaled fp8 safetensor clip/t5 atm. Just wait some :) Just tested it and worked perfect. Civitai has problems with uploads... still uploading ...
darksidewalker darksidewalker 10 hours ago you posted this, yet I still don't see them. The state of this site is unfortunate on several levels. I really like it a lot, but between technical issues and censorship requirements for payment processors... I mean, it's a site focused on AI, can they not use coding AI to make things work better? Legal AI to work around the payment processors? Sorry for the rant.
rando2048 Maybe try to delete the browser cache, all is uploaded and can be found by hash, too. Hope that helps!
"This version is missing files"
Reuploading. I deleted it by accident :(
Details
Files
wan21Lightspeed_lightspeedI2v14B480p.safetensors
Mirrors
wan21Lightspeed_lightspeedI2v14B480p.safetensors
wan21Lightspeed_lightspeedI2v14B480p.safetensors
wan21Lightspeed_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
dasiwaWan21_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
Mirrors
wan21Lightspeed_lightspeedI2v14B480p.safetensors
wan21Lightspeed_lightspeedI2v14B480p.safetensors
wan21Lightspeed_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
dasiwaWAN21_lightspeedI2v14B480p.safetensors
dasiwaWan21_lightspeedI2v14B480p.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.