Trained with OSTRIS [email protected]
V2 Update:
The V2 LoRas are trained on a single Double Transformer Block, which should result in less bleeding. Despite being trained with a rank of 32, the LoRas are only 9.4 megabytes in size.
V1:
This is a collection of various classic to modern photo films.
This is the first version. I'm already working on a second version, where I'm trying to pinpoint which single transformer blocks hold the style information. Once I've figured this out, I can train exactly those blocks to create even more flexible and, most importantly, much smaller Loras.
This will also make it possible to quickly train additional films.
Enjoy the Lora!
Description
FAQ
Comments (2)
Great idea! Doesn't seem to work too well in Forge, even with no other Loras. Seems to need Distilled Cfg (Guidance cfg in comfy) = 1 to produce tmax as monochrome, otherwise its color...
I did not test it with forge because I m only using comfyui. Is it just the v2 LoRa or also the V1 ?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.