Note: work in progress
humana tries to focus on natural looking depictions of people; which is in contrast to opticon, which aims for striking and stunning images.
I strongly recommend using DPM++ SDE Karras
If you're aiming for authentic looking images, some guidance:
Include candid in the prompt.
If you want selfie style photos, selfie pov includes a phone in the image less often compared to just selfie.
Including depth of field, bokeh often improves visual quality in general.
Description
FAQ
Comments (24)
This is an exceptional start. I'd like to see it further developed. Sent you some buzz as an incentive :P
Thank you! Will keep working on it - it would be very helpful to get any feedback on what areas you think needs most work and what directions you believe would improve it the most
I think what I'm mostly noticing is that it's quite different in prompt adherence when compared with others (the problem is that there's a "sameness" with the others which is exactly why I like humana). Maybe it's just a matter of my learning its language. Would it be possible to generate some more examples on the model page showing things like angle, pov, numbers of people, nationality, etc. since you know the language it was trained on?
I should mention, and I know it can be a pain, but not having the prompts in your example images makes this much harder :P
@sarashinai I'll get right on that; good idea!
@sarashinai I've tried to upload some examples; please let me know if it's at all what you had in mind
@sarashinai I'll also post here that I'm looking at improving facial variety and the Turbo lineage of models seems to perform really well https://civitai.com/posts/8116046
Here's same prompt random seeds with my turbo merge https://civitai.com/posts/8147963
@epiTune That's a ton of examples, give me some time to work through them, thank you!
@epiTune Those look great! Turbo lineage not out yet, right?
@sarashinai I'll post the turbo model tomorrow
@epiTune The faces one was a great help both for different tones but also for the prompting in general. I can't seem to get that shimmering skin colour that yours show, though.
@sarashinai depth of field, bokeh, analog photo was the most common through-line in the prompts I uploaded
@epiTune All well with turbo? With you?
@sarashinai Hope to share more updates on Turbo and other models soon!
@sarashinai You might want to check out https://civitai.com/models/934199/portaito
Indeed the most realistic pony. But you need to find the balance between realism and flexibility.
Because the too much realism the lora compability, likeness is poor. I speak about celebs, characters.
Im waiting the next versions, thanx.
Thank you so much for taking the time to give feedback!
Is what you're saying that this model struggles to work well with LoRAs due to what you believe is a mismatch in realism? Would it be possible for you to mention some LoRAs you found worked poorly?
@epiTune Hi! The 90 percent of Pony characters and celeb loras made on semireal, outdated checkpoints. Those checkpoint, which are balanced and kept their anime roots are compatible with all the loras, and the likeness is very good. You put too much realism, which sucked out the likeness(i cant express myself better).
Check out the latest Cyber pony models, which are very balanced with good, but not great realism.
Your model is the best at realism, but who wants to create always 1girl? Answering your guestion: not a single celeb lora has the likeness, they are like 1girl.
Anyway its very good and has effortless realism without ton of negatives and quality tags. Im curious, how is this a Pony model at all. It could be a realistic XL model.
Good job. Wish good progress to you.
@kolevupu586 If you want a model that is decent at realism but handles what you call "Pony" roots better I'd recommend opticon
Having better face variety to get away from samey 1girl looks is something I'm working on - https://civitai.com/posts/8116046
why are the prompts not visible for your example images? i often like to view those to see where I should start
The tool I use for image generation doesn't add the prompt to the metadata of the image
WOW...WOW...WOW!!!
LOVE ALL YOUR Models
THANKS!!!
Excellent model! You really should do a version 2 because I've never seen a model do such realistic humans without loras before. Great job!
Thank you! If you haven't tried it I can recommend giving portaito a go https://civitai.com/models/934199/portaito
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.















