Wan2.2 SmoothMix T2V Fan Made Merge | Recommended LightX2V LoRAs H | L or SEKO
A balanced and refined merge focused on better anatomy control while keeping the smooth aesthetic of the SmoothMix series.
Recommended Settings
Sampler: DPM++ 2M SDE / Beta Steps: 6
Sampler: Euler / Beta Steps: 4
Important: You need to use the LightX2V LoRA with this model if 4 steps is your thing.
Key Improvements
Slimmer and more natural female body (prompt keywords have strong influence)
Improved male genitalia (trained on a dedicated 100% male only dataset)
Better overall LoRA balance (merged at 10-23% below sweet spot)
Reduced unwanted "bounces"
Why This Merge Exists
As a fan of the SmoothMix series, I created this merge to solve some issues I kept running into with the v3 T2V model. I trained two custom LoRAs: one for realistic male genitalia and one for a slimmer female physique.
The final merge is intentionally tuned with slightly lowered LoRA weights. This makes the model much more stable.
The base model + the flat chest LoRA could produce 100% flat-chested women. After merging with SmoothMix, completely flat chests are no longer the default. However, specific prompt keywords can still bring back a flatter look. The stronger stylistic influence of SmoothMix tends to favor larg breasts instead of small or flat chests.
Pro Tip
Generate 3 videos. If you don’t like the result , say "fu*************k" and delete the model.
Description
FAQ
Comments (3)
I'd be curious to try your LoRA's you trained for this outside of the checkpoint. Consider posting them.
Works great with the lightspeed lora which is what I'm trying it with as it mimics smoothmix. What CFG and how many steps to run it normally though?
good to hear :), Well, it depends on the prompt. The best way is to set the video length to 1 frame and render the first frame (usually 720x1280 gives a good preview). From there, you can adjust the settings.
High noise is the main architecture -> it works with what it receives from the text encoder to generate an image that follows your prompt closely.
Low noise acts as the refiner and detailer.
(keep in mind that UM T5 XXL is quite limited with NSFW content. most of the time the problem comes from the text encoder, not the WAN model itself.)
So if the 1-frame result isn’t close to what you want, focus on improving the text encoder (cum => translucent liquid substance, Blood => thick red fluid substance etc.) and high noise settings. If the image is already close to your prompt, then switch to low noise and fine-tune the settings for better quality and details. (3 seconds in 512x512 is fast and gives a good overall preview )
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.