Trained on 200+ very highly detailed high resolution (min 2000px) images + 35 hires vidclips of amateur female face portraits, with visible skin pores, skin hairs and blemishes/moles.
The lora is intended to use for higher resolutions like 768px and up (width and/or height) - as training concentrated image and vidframes samples between 768 and 1152px this is where performs best.
My samples posted are made with DPM++2M, Scheduler Beta for 28 steps on the original tencent bf16 models at 512x768 resolution @73 frames, lora strength 0.85. The quants/fpX/fastvid I tried all produced visible different and degraded quality IMHO.
V1.1 works better with other Loras - put it around 3/4 weight of the other Lora; e.g. this Lora at 0.65, the other Lora (character etc.) at 0.9
For SkyReels or Hunyuan ImageToVideo vids, lower the strength to about 0.4 for vids over 73 frames. the Lora was only trained on snippets <73 frames, and this seems to have more visible (anti)effect on longer vids based on I2V model than T2V.
Description
FAQ
Comments (9)
Super detailed ! not only the skin but the teeth! very neat training
i am happy my imageset consists of a lot of great quality teeth images as well, yes its very important for realistic renders to get teeth right, glad you like it!
Hello, thanks for your contribution. I have discovered this model is also good for making any kind of realistic person video, since the base model is prone to generating "CGI" or "Pony" people, this model helps remove that and always generates "real" looking people.
great to hear! can you share any vid you made to illustrate I am curious
@jhnprst2150 I just posted an example
thanks for that; did you also compare to same vid without this lora? i want to move to next version to include more 'faraway' shots not just closeups, so its nice to hear already good feedback
@jhnprst2150 Thank you! I did not explicitly test this video but have been extensively testing Hunyuan since release date and I can tell you with 100% certainty this LORA makes the videos look more real than base model (probably at the cost of not being able to generate Anime or 3d renders or whatever, but that's not my use case). I'd love to test your new, more generalized version once you release it, I will definitely be on the lookout.
thx to you both, shared my very special workflow here: https://civitai.com/images/49712190
somehow your lora is "unlocking" some kind of realism. really see forward on @jhnprst2150 next version. @AIVideoEnthusiast
@TedBerlim thank you for the tip and posting the worklow, it looks way more advanced than my basic one, I am going to study it!! and yes i am working on a next version, lets keep in touch!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.