The embedding were trained using A1111 TI for the 768px Stable Diffusion v2.0 model. The embedding should work on any model that uses SD v2.0 as a base.
Usage for A1111 WebUI
Download the TungstenDispo.pt file and put in embeddings/. Prepend "TungstenDispo" at start of prompt.
V1:
A total of ~100 training images of tungsten photographs taken with CineStill 800T were used. The split was around 50/50 people landscapes.
The effect isn't quite the tungsten photo effect I was going for, but creates very nice, artistic portraits of people. For some of the people, I used SoCalGuitarist's Negative FaceLift as a negative embedding.
I used it on 0.3 strength, and it seems like it makes the eyes slightly less wonky. Unclear extent of effect.
Landscapes haven't been experimented with much and are WIP.
Description
Workflow for Above Pictures
Sampler: Euler-A, 20 Steps, CFG: 7.0. Slightly cherry-picked for best pictures.
900x768 -> 4x LDSR upscaled
Negative Prompt for all images (Not entirely sure if all of them matter, but does help a bit):
> (Neg_Facelift768:0.3), (blur:0.3), (cropped:1.3), (ugly:1.3), (bad anatomy:1.2), (disfigured:1.1), (deformed:1.1), (bad proportions:1.3), (extra limbs:1.2), (missing fingers:1.2), (extra fingers:1.2), (out of frame:1.3), (makeup:1.1)
Positive Prompts:
First Image:
>TungstenDispo, photoshoot of a asian female model with white hair, in a dark room, (closeup:0.2)
Second Image:
>(TungstenDispo:1.3), photoshoot of a (model:0.7), posed, in a dark room, highly detailed, (closeup:0.2), (skin pores:0.5)
Third Image:
>(TungstenDispo:1.2), photoshoot of a model, posed, in a dark room, (closeup:0.2)
Fourth Image:
>TungstenDispo, photoshoot of a male model, posed, in a dark room, (closeup:0.2)
FAQ
Comments (4)
Is it possible to use it in v1.5? The WebUI doesn't seems to detect the embedding at all :(
Hi! Yeah this is trained on v2.1 in an effort to modernize :( sorry about that. I recommend using analog diffusion and lots of prompt tuning if you want to replicate a similar style with 1.5 (https://huggingface.co/wavymulder/Analog-Diffusion)
@fzbuzz Bummer! But thanks for the hint I'll check it out ;-)
In InvokeAI, I found that I needed to use the trigger <CineStyle> rather than the filename.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.