a lora for minimaxh3 ...

@kimvermeersch231 Sorry. I'm really playing around with this Lora in my head, and it's amazing. At a strength of 0.75, it completely breaks the “plastic effect” for me.

So: Sorry. I really didn't have the patience for the question above earlier.

@igotothebeach8973 So. Now about you—sorry—you caught me in a really bad mood earlier. Let me explain something: the FP100C is part of the instant film family. I’d actually been using it way too much because it can reproduce extremely natural skin tones. It has a second advantage, which is that you get a color negative of the highest quality in roughly double medium format. I processed the Lora for two reasons: extreme boredom and the idea of breaking away from that damn plastic look. If you want to experiment: I currently have absolutely no idea about Minimax Lora processing, but this “beast” here produces extremely good results at a density of 0.75.

If you happen to catch me in a good mood, I’ll dig out my old Kodachrome, Orwo (GDR), or Russian Svema films in the next few days—and your question was just too vague. I couldn’t make head or tail of it.

Best regards.

@misterpurplepill wow. try strength 0.7 - 0.75

I’d be more than happy to fulfil your request straight away. The only problem is that I don’t have any clean raw data for the training at all. I’m also currently churning out videos like mad with the Lora on my local machine. I never would have thought that the skin tones and the ‘feel’ would be so striking. But… I’ve no idea where I can get the raw data for these old smartphone recordings. Maybe I’ll shoot something over the next few days with overlaid Kodachrome and Kodak films – they come very close to that ‘era’.

@aixarts You might find something on YouTube, and in that case, I'd extract the frames from the videos. However, I'm very particular about raw data: when I work with LORAs like this, I want to train them on the actual hardware, so I need to be absolutely certain that the raw data actually comes from that hardware.

Sample: "fp_100c film style, cinematic medium two-shot on the sun-bleached travertine terrace of a modernist hilltop villa overlooking the Mediterranean coastline at late afternoon.

Blazing high-angle sun creates hard graphic shadows beneath a broad canvas awning. Leaning against the honed stone balustrade with an espresso is <Picture 1> Johannes (wearing a light sand-colored tailored linen shirt with rolled-up sleeves, calm weathered gaze looking out toward the sea).

Relaxing nearby in a low teak lounge chair is <Picture 2> Anna (wearing an olive silk summer dress, dark hair swept back). She wears dark vintage tortoiseshell sunglasses; with a slow, deliberate gesture, she slides the glasses slightly down the bridge of her nose with two fingers, her sharp, intelligent dark eyes locking directly onto Johannes in the bright daylight. A warm coastal breeze flutters the linen awning.

STRICT CLOSED-MOUTH LOCK: All lips remain tightly shut, relaxed natural breath, zero speech.

Legendary fujifilm fp100c sunlight latitude, non-clipped skin highlights, saturated turquoise sea, rich terracotta and stone earth tones, organic silver-halide grain, shot on 50mm f/4.0, slow cinematic horizontal pan, hyperrealistic 4k."

I'm currently trying to get “ostris” to run on my machine, so I'm a little stressed. Could you explain “workflow” in more detail? So, I trained this LORA using Civitai. Is that what you mean? Are you referring to specific prompts?

Feel free to ask questions. Training LORAs via Civitai is extremely simple, but I want to figure this out offline as well, and once I’ve managed that, I’ll have a few more “analog LORAs” up and running very quickly.

@vastsun78523 Well, I trained Lora with Civitai because my hardware isn't powerful enough for training VideoLora. But for my videos—including promotional videos—I always use Maestro. I actually recommend Maestro because you don't have to deal with that workflow logic.

I’m currently using the FP100C for image videos related to the real estate industry, and I still have a few “training videos” in the pipeline where so-called “person-specific services” are an important topic. If you’re experimenting, try strength 0.7 or 0.75. And here’s another tip I can recommend: use a prompt to simulate the “Helios lens.”