Aslan's Laconia - Ancient Rome
These embeddings are trained on movies set in Ancient Greece and Rome from 50s to the 2010s. I ultimately wanted something that could combine the classic Gladiator aesthetic and the realism of live action cinematography.
--------
To use these embeddings in AUTOMATIC1111, copy them to your embeddings folder in your stable-diffusion-webui directory. No need to restart, just invoke the embedding from your prompt. Make sure you are using a compatible model or you will get errors!
--------
The embedding was trained on 351 768x768 images that were cropped and tagged using BLIP for training then manually corrected. (This is a painstaking process)
--------
Like most other 2.1 textual inversions this works best when paired up with other embeddings and good negative prompting. Check my examples for ideas.
Description
This version will not be the last. I want more coherence on the armor and togas. But the settings and architecture are where I want them.
Batch size of 10 and Gradient accumulation of 36.
Learning rate: 0.05:10, 0.02:20, 0.01:60, 0.005:200, 0.002:500, 0.001:3000, 0.0005
FAQ
Comments (2)
This is beautiful! Really feels like you unearthed some old films.
Looks like face restoration is disrupting visual continuity - making faces that would have been rendered a little out-of-focus as too sharp instead. Better to accept some bad faces and revisit them with inpainting after. With the cinematic training behind this embedding I expect it'll generate a lot of that short depth of field.
I appreciate it! It does a great job of giving shallow dof and generally following the rule of thirds with framing. As far as faces goes the best solution I have found in practice is to use the facelift-neg embedding and Hi-res fix for anything between a mid-wide to mid-close shot and for wide angle to just hope for the best and use inpainting or other programs. That said I left in my less desirable results in the samples to give a more realistic idea of what to expect. Thanks for giving your thoughts!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.