Trigger phrase: Japanese color film photographA low rank test with a rather unrefined dataset I decided to upload as it turned out better than expected. Still doesn't quite look the way I would want.Tested on SwarmUI and Forge Neo, with Z-Image Base. All images posted are raw, with prompts and metadata visible, and no other LoRAs used except Turbo.Model showcase gallery: Forge Neo, 8 steps, 8 step Turbo LoRA, LCM + FlowMatchEulerDiscrete, high-ish resolution and no upscale. Just works.Post after that: SwarmUI, 8 steps, 4 step Turbo LoRA @ 0.5 strength, Euler A + Beta, standard XL resolution and 1.5-2x upscale. Good and fast but a bit desaturated.Very briefly tested with Z-Image Turbo, effect seemed low. I don't really like ZiT, and may not create a specific version for it. Not tested on the Civitai generator.I have various plans for these LoRAs like fully separating out the mid-to-late eighties from the nineties, and sfw idol content from nsfw erotica content, but couldn't predict when or if any will be realized. I've also been impressed by Krea 2 lately like everyone else I guess, and may swerve that way.For anything else, description of the Flux version <a target="_blank" rel="ugc" href="https://civarchive.com/models/1978169">https://civarchive.com/models/1978169</a> should mostly apply.

Files with SHA256: 41e2ae20040ef929602ccdc523d9288bfada8ec3052b3c7f51d7a6882d5e9c96