A number of models exist both as models and as LoRA-s.
| Model | Notes | Preferences |
|---|---|---|
| Dyno | T2V Wan 2.2 High | Better as full-fat than as LoRa |
| Wan 2.2 I2V Moe Distill Lightx2v | I2V Wan 2.2 High only | As good as a LoRa? |
| CausVid | Wan 2.1, can be used as Wan 2.2 Low | Better as full-fat |
| Krea “Realtime” | Distill of Wan 2.1 14B T2V | Definitely better as full-fat |
| VACE 2.1 | VACE 2.1 was a full T2V derivative 27Gb in size before it was extracted | It is widely believed it works just as well as extracted layers |
Doesn’t work as a LoRa, somewhat works as a full-fat model. Difficult to use. Produces naturally looking vidoes: soft colors, normal contrast. Has been used instead of high-noise in Wan 2.2 workflow with rCM lora on high.
Finally something that looks kind of realistic.. other 4 step models look so artificial
krea is causal like causvid