penis on face

This model is based on VAM T2V and was originally trained for text-to-video generation.

While it works in standard T2V workflows, results can be inconsistent depending on prompt structure and motion complexity.

For more stable and controlled outputs, I recommend using it in an Image-to-Video (I2V) workflow. It performs especially well in Wan 2.2 I2V, where motion handling and conditioning seem to align better with this model’s training style.

If you’re getting unstable motion, warped frames, or weak prompt adherence in T2V, try switching to I2V with a strong reference frame. The difference is noticeable.

Mode -----------Works--------- Recommended ?

wanT2V---------- yes ------------sometimes

wanI2V----------- yes--------------- good

wan2.2 I2V--------yes-------------- Best