This is a LoRA that's trained on retro CGI images from the late 90s and early 2000s meant to reproduce the slightly surreal look of the renders of those times.
The model uses "rtrcgi" as a trigger word at the beginning of a prompt. It was trained on the base models and generalizes well over different checkpoints and in combination with other loras. However, you might need to fiddle around with the strength of the trigger word and the LoRA model weights themselves to get a good result.
I've noticed some problems when the prompt gets too complicated and the scene too complex. I might come back to it in the future to try with an improved training set but for now the quality is good enough to be usable for interesting generations.
The SD1.5 LoRA is generally a little less fiddly and easier to use than the SDXL version.
Please note: the LoRAs were trained on 512x512 resolution images, that includes the SDXL LoRA. From my experience it is better to generate at 512x512 or 768x768 due to this limitation. Other resolutions also work, but some artifacts might occur.
Description
FAQ
Comments (3)
I wonder what it'll do if used as a negative
I actually already tried that a few days ago already but the results were inconclusive.
With cartoony generations and other less realistic images, using this model as a negative will push those generations towards more realistic images - less smooth skin, more realistic lighting etc.
With generations that are realistic already, it only seems to shuffle around some details. At least I've not seen it do much to those images.
So if you're struggling to get rid of a cartoony/computer generated look in some of your generations, it could be worth it using this lora as a negative but in most othrt situations it doesn't really change much from my experience.
It'd amazing
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.