phenidone is an early work in progress photorealistic Illustrious checkpoint.
Compared to anime Illustrious checkpoints, you should expect it to:
Handle high resolutions worse
Have worse consistency relating to composition and anatomy
Compared to photorealistic Pony checkpoints, you should expect it to:
Have worse skin-texture
Have worse background and scene coherence and detail
But in general, you can look forward to a breath of fresh air; and hopefully you can have fun experimenting with it.
Settings I use:
832x1216 and 1024x1496, 30 Steps, Euler A, 5 - 12 cfg
I recommend trying varying cfg based on what results you're after; here are examples of 5, 7, and 12 cfg.
Grids generated with these prompt styles,
BASIC PHOTOGRAPHY
depth of field, bokeh, photo, incredibly absurdres, highres, masterpiece, newest
DRAMATIC PHOTOGRAPHY
colorful, saturated, high contrast, bright lighting, photo, incredibly absurdres, highres, masterpiece, newest
MOODY B&W PHOTOGRAPHY
monochrome, high contrast, harsh lighting, photo, incredibly absurdres, highres, masterpiece, newest
SURREAL
surreal, colorful, photo, incredibly absurdres, highres, masterpiece, newest
All images in showcase and that I upload to gallery are directly from the checkpoint with the settings above (5cfg) - no post-processing.
Description
FAQ
Comments (14)
Girls look very similar to each other. Not sure how to fix this.
Avoid words like, "1girl," and instead use, "unique face," "girl next door," or a nationality such as "British," etc. You can also try a random name like, "Gwen," etc. 1girl as a word tends to filter the generations down to a single style and look that's fairly homogenous, which is an issue I find with most models. I hope this helps mate.
@Shio_N phenidone is still early work in progress, eventually I'll use techniques similar to what I used to increase face variance in portaito https://civitai.com/models/934199/portaito
I think EpiTune is onto something here. I found that using this model on the backend to run 80% of the generation that the model really shined. On a whim, I grabbed Epi's Glances as an initial 20% base to help kick the model with a posture and expression, and Phenidone did the rest. I spent a little time inpainting, and with a bit of tuning was able to get stunning results.
I think with some work, Epi's got a winner here that could rival the absolute best on Civitai. Detail, vividness, fantastic skin and faces... It's all there in the bones.
Holy sh*t, just learned something new. Remixed an image with your 'workflow' and used 'glances' as 20% base and 'phenidone' as 80% refiner, and the result was amazing... o.O
Original without 'refining': https://civitai.com/images/60111851
Remix with 'refining': https://civitai.com/images/60115987
@Lianis this is one of those dirty tricks I've been using for my images for a while now. I sometimes start with a base anime model to kick the photoreal models into a unique picture. It's silly, but I think it produces excellent results. :)
@Lianis @TurinBjorn I concur with the holy sh*t! This has opened up a whole new world of experimenting. Hello sleepless nights...
@epiTune How's this one do for long prompts?
Really don't find any fun in doing short one-sentence prompts, which I've noticed you've been using for the examples in your last few models.
Another great model by epiTune! It's just hard to convince to generate backgrounds when used together with LORAs. I am putting a lot of weight for that in my gens to compensate.
baby你的写实模型就这个最棒了身体纹理清晰,还有新的模型吗宝
its different, that counts as a +1 for me xD , i would like a version 2.0