Best settings (My findings)
Been messing with the model for 2 months having great fun. Here are some go to settings i've found work quite well and are very different from what i've found so wanted to share with others if it's any use.
How i'm using it:
With ComfyUI, "Wan2_2-TI2V-5B-Turbo-Q5_K_M", Primarily for I2V but works just fine for T2V also
lora strength (Turbo_lora_rank_64_fp16): -0.2
SamplingSD3 shift: 6 - 18 (above 23 or so starts causing issues, below 5 causes visual artefacts)
With the shift value I found staying around 8-12 gives best results but varies depending on desired motion
Steps: 5
CFG: 1.0
Sampler: euler (uni-pc is suggested but after testing euler gave more realistic identity and object shape)
Sheduler: simple
LatentMultiply: 0.85
Resolution: 768x1024, 1024x768 or above, (576x768 is useable but reducing quality a lot)
length: 41-121 (I found 69 gives me good results on my limited hardware)
Hope these settings help anyone wanting to use a video generation model without having high specs, this works pretty well on 6GB Vram (RTX 3060 mobile)