Hugging Face's logo

Phr00t
/
LTX2-Rapid-Merges

Image-Text-to-Video
ltx2
t2v
i2v
Model card Files
xet
Community
92

Terrible video quality and consistency for some reason

#90
by Kamillia - opened
Discussion
Kamillia

I'm using the LTXV-DoAlmostEverything-v3.json workflow file with ltx-2-19b-phr00tmerge-nsfw-v6.safetensors.
For some reasons the videos come out looking like Will Smith eating spaghetti in 20023. Faces are unrecognizable, characters are morphing and mutating. Utterly terrible.
Any idea what the problem might be here? I didn't touch any of the settings.

I had great successes with Phr00t's WAN2.2-14B-Rapid-AllInOne in the past, and I was hoping this one would even better.
Diosis

Hi,

  1. Try changing the duration, FPS above 16, RIFE, resolution should be a multiple of 64.
  2. This model is very sensitive to prompts. For best results, create prompts according to the recommendations here: https://ltx.io/model/model-blog/prompting-guide-for-ltx-2
  3. If you're using a fixed seed, try changing it to random.
  4. Try V62.
  5. Try changing the workflow and adding a negative prompt.

Sign up or log in to comment