This series of the models was started (v0.4) from DucHaiten-Pony-Real (beta) by DucHaiten.
Version v1.0 was trained using a LoRa mode in One trainer. Next version are trained in Fine Tuning mode (full SDXL training).
Up to v1.0: Photorealism, lighting and anatomy related LoRas from CinEro series are integrated into this model in order to amplify cinematic render.
Goal of training was to improve to stability of anatomy and make the picture more realistic and moody / cinematic.
Version 1.1: shifted to fantasy / sci-fi themes with less realistic and more contrast / unstable rendering.
Version 1.2 [SOON]: is v1.1 further development with shift to horror / tentacles theme.
PS: be responsible and don't abuse this unrestricted model.
Description
Integrates my latest trainings with landscapes and portraits. Tuned to give more smaller details and micro-contrast. Adds vibrant colors, intricate details, and creativity in general.
FAQ
Comments (19)
how can I increase the depth of field because I find there's too much blur in my images?
Is the V09dFp16 version better than the other versions?
Thank you
Very good model, otherwise!
Adding "shallow depth of field" into negative should phase down the bokeh effect. Also "shallow dof" in negative can do the job. Additional negative triggers for you: blurry background, out of focus.
Also, try adding the "sharp focus".
Later I can test if triggers like "f/8" or "f8.0" can work (f4.0, f5.6, f8.0, f16.0). These photographic triggers are from SDXL, but who knows if they sneaked into the Pony as well.
As for versions, I don't release new version if it don't satisfies me more than previous in some of the aspects.
0.9d is was good for me 'cause I love bokeh. But if I'm not mistaken, some users reported that it has too much bokeh.
https://civitai.com/images/23329550
look at this image above. It is v1.0 with "shallow dof" and "crispy details" in positive. And it has a depth of field more shallow. Maybe you can peek up some good prompting from it.
https://civitai.com/images/24251497
This is also an example of wider field of sharpness.
"mobile photo" and "shot on smartphone camera" - in past I was using these triggers as negatives to make a picture more photographic. You can do the opposite.
@homoludens OK, thanks for all this information, I'll try it out later.
another question: have you integrated specific cinematographic styles, like Christopher Nolan, Ridley Scott, etc.?
or famous photographers?
@Suzanne Nope. I didn't find a good one which works for me. And honestly didn't test those properly.
AND I don't expect it possible to combine all distinct styles without messing up. I decided to integrate a style from my own photographs. For example Meostigmat Lens style. In future I want to take a photographs on other vintage lenses with cinematic colors: one lens analogous to Carl Zeiss Sonar + Helios 44-2 + More photos from Meostigmat.
@homoludens Thanks for these details, I didn't know you were a photographer.
I would like to do more studio nude images in the professional style, but it's difficult to find the right models or the right loras.
@Suzanne I would strongly recommend you to learn ComfyUI.
There are custom nodes applying the HaldCLUT effects.
http://rawpedia.rawtherapee.com/Film_Simulation
In my experience, it is MUCH MUCH more effective and precise to apply the text and technical (dumb) effects using a special node instead of trying to find a LoRa giving the stable result you imagine in at least 75% of runs. Don't waste your time trying to apply color and light grading with LORA. Just install a light and color modification nodes into ComfyUI. There are also a film grain effects, etc.
With ComfyUI you can draw your subject on white or other simple background, create a mask from text prompt (Masquerade Nodes) and then cut and paste the subject onto another image with consistent background. Masquerade node (Text to Mask) can almost select anything matching the prompt, but it eats VRAM. It can be replaced with a specialized masking nodes who knows only human.
So, in ComfyUI you can mix a photo of your studio with rendered subject and vice versa.
A1111 WebUI is good for finalization with Inpainting.
@homoludens Well, I use Fooocus because I like its simplicity and speed, and ComfyUI seems a bit complicated to me, but it's certainly more powerful and I only have 8GB of Vram and 32GB of ram - is that enough?
Thanks for all your replies
@Suzanne In ComfyUI I've used the SDXL with Nvidia 3050 Ti 8G VRAM with quite high resolution (more than 1024). In Foocus you should also be able to work with 8G VRAM.
For the A1111 WebUI 8G is quite low.
I don't know if Foocus supports any extension. If it do support extensions then try to find some color / light grading extensions instead of LORA. Because exposure modification is a quite simple algorithm in comparison with LORA. Why would one develop a highly sophisticated android to just scratch his/her head?
@Suzanne You can past any of these images into the ComfyUI to get a see the CLUTs in action.
These images has a workflow with FLUX Dev quantized model, upscaler and CLUTs.
https://civitai.com/posts/5583506?returnUrl=%2Fmodels%2F647237%3FmodelVersionId%3D724099
@homoludens
The word "ART" has been added to the version name since v1.1.Does this mean that it is less realistic and more artistic than v1.0?
"More creative" I'd say. And due to increased contrast, realism was a bit suppressed (little bit in some cases). Also, eyes and some other face features can be rendered less realistic.
This version had a similar goal as a "Spicy" version of CinEro SDXL
@homoludens
Thank you for letting me know.
Hmmmm Not sure if I get this message 100% correct...
Is it request to train FLUX to get similar horny stuff as PONY models?
@homoludens YEEESSS!!! Is it possible ?