Thank you!

#149
by vasilmitov - opened

Thank you for your hard job and for the sharing of this and the other models in such a simplicity way, it feels like 'retro' sd/xl :)

I will be happy to help with anything - testing staff or whatever, and i can give results in any format, or do whatever to help with your project...
... basic editing skills on most languages
.. expert video/photo/lighting edit, record, hardware...
... i believe where around September 2022 i started comfy and never quit, so i know this and that...
NVIDIA GeForce RTX 4060 Ti - 16gb vram / 32gb ram / win 11

sorry if this is not the right place of offering help, as i said - i am more of a graphic person

I do know someone was asking about what keywords can trigger certain behavior.

  1. Research all the LoRAs added (from the metadata seen on Hugging Face on the file) and create a document that includes all the key words and phrases. Also include origin URL. I haven't been able to find all of them, but I think they are all on civitai. (Civitai drives me nuts as they hide the model filename unless you click download.)
  2. Camera movement guide. This might be limited due to the following issue:
  3. Find a way to combat slow-motion videos. For instance, if you try to make someone surf or snow ski, it's all slow motion. I think it is because there is a lack of high-noise models and LoRAs, but I haven't had the time to recreate Rapid and experiment by adding high-noise or create a Rapid High version that would run for the first 2 steps then switch to regular Rapid (Low) for the remaining steps (3-4).
  4. Find a good way to add voice+lip sync into Rapid. All I have is using a Rapid start frame and then put it into an Ovi workflow. Basically just doing I2VA with Ovi. Ovi's video model is not straight forward. I was hoping to do some sort of model merge with Rapid or at least add WAN 2.x LoRAs, but the blocks don't align.

Doing it, will come back soon.

I am going to use "wan2.2-rapid-mega-aio-v7.safetensors" as a base... if you recomm any other as a base correct me

Sign up or log in to comment