W25 Realistic is an experimental LoRA designed to enrich and transform the visual language of WAN 2.2 — both for image-to-video and text-to-video generation — pushing it closer to the look and feel showcased in the WAN 2.5 preview.
⚠️ Important: This is not an official model, nor a distilled version of WAN 2.5. The dataset does not originate entirely from 2.5 outputs; instead, it is inspired by them. It combines a heterogeneous mix of sources, including highly realistic and cinematic clips curated to capture the emotional depth and visual intensity seen in the 2.5 previews.
By applying this LoRA, WAN 2.2 generations tend to become:
🎥 More cinematic: richer composition, improved lighting behavior, and more nuanced camera dynamics.
❤️ More emotional: heightened atmosphere, better storytelling tone, and expressive visual language.
Description
FAQ
Comments (15)
Did you train using videos as well or just images?
just videos 👌🙌
@Ciro_Negrogni Great work! Congrats!
@rafaaguiar990170 🙏🙌 thank you very much
i'd love a "vs" with / without lora :)
what exactly is this lol?
Videos of WAN 2.5 trained on 2.2
@LDWorksDavid and prompts from higgsfield
@fauxtographypro The dataset also contains other realistic and cinematic clips that were reminiscent of the style of the 2.5 preview results. in general to emphasize both emotionality and richness in detail
so what we just drop this in and everything looks better now in 2.2?
@jac001 In a different aesthetic more than "better".
Some marketing BS
@Ciro_Negrogni I would recommend using the LORA ON / LORA OFF post as the primary preview and mentioning the increase in expressivity as the main feature in the description. I really had to do some digging to see the value of this lora and that post is what convinced me.
Otherwise, I'd say you aren't likely to see much interest in this because as-is it is difficult to understand what this even is, as the primary comment accentuates.
@Clocksmith Maybe now it's a little clearer, thanks for the advice