Welcome to the Everything Looks Like A (ELLA) Model
technically the first release is a tuned Photo-Model based on my epiCRealism XL CC model
Photo V1 - should generate better authentic images overall
Description
FAQ
Comments (25)
A-Class
Any chance you'll be trying out Chroma or Qwen?
Since I’m not a fan of comfyui and the weight of that models I have no plans for that. Also these models are very powerful on its own, so there is not much room for improvement and also there a pretty good Lora’s around already.
@epinikion Chroma's default photorealism is FLUX-alike due its engine. LORAs help but can't stop Chroma's love to fall into anime.
@mphobbit Chroma looks less sloppy than Flux, use a normal realistic lora, there are some options. Also, someone else will finetune it
@epinikion I can't blame you there! I'm not a fan of comfyui either but sadly the gradio based systems I liked more, suck as Forge, are dying off. I've begrudgingly been using comfy because models like Chroma stand above anything SDXL, especially when it comes to more complex prompt adherence.
@2P2 If Chroma wants to transform picture to anime, it'll do it despte LORA and prompting tricks (especially she likes convert sci fi to anime). With photorealisitc LORA the image will be something like "anime through filter". I know what to do but I'd appreciate any Chroma-based checkpoint tuned towards photorealism.
Chroma was initially positioned as a raw material for finely tuned models. In its original form it is unstable and inconsistent. This would be a great model if someone stabilized it towards NSFW photorealism
This new model is a big change compared to the epiCRealism XL series style. The details are quite good, but I found a serious problem. Almost all indoor scenes have a flash effect.
Need a bit more testing on this. Will see what the community will come up with to see the flaws in this.
I get that flash effect (harsh shadows immediately behind the subject) no matter what I use. I wish I knew how to avoid it but, at the moment, I am rather avoiding indoor scenes.
up until now the flash effect didn't really bother me, but now that you mention it - The checkpoint seems to ignore all prompts that include lighting setups in an indoor setup (e.g. Rembrandt lightings, chiaroscuro , 3 point lighting etc..(, and also any lora that I've tested that has to do with lighting. Would be fantastic if you could address these aspects and still deliver what you've hinted! :-)
Try on positive:
(no flash photography:1.5), (no flood light:1.5)
Variations also work, use related terms to specify scene lightning. Using lightning LORAs also does the trick.
Posted an image with a misplaced eye anyway :( poor woman
https://civit.ai/images/106326913
yah so... like an actual human? "icky" asymmetry?
I am excited for the news :)
Impressed! - and also looking forward to next week! Your work is constantly awesome...
Ty
I'm your biggest fan epinikion!!!
Are you?
@epinikion yes!!!
well I'm 6'6. I could be the biggest
sad to see it being deleted soon.. Any chance you will upload somewhere else?
And out of curiosity what is changing in terms?
I will let it here i guess but will not push it into generation service anymore :)
good!