Panima

Use https://civitai.red/models/2560840/anima-turbo-lora with the model if you want similar output. Use 8-12 steps, more steps usually giving more crips images. You can probably get the same without but I have no idea about the negatives you would need. Heun + beta is so far the best combination of sampler and scheduler I have found, creating the most lifelike images.

This model is a modest realistic finetune with Anima, which seems to be surprisingly good at learning the photographic rendition. The v0.5 has been trained for 50 epochs on same dataset as LoZ, balanced slightly. The v1.0 is full 100 epochs, with the last 50 epochs with smaller portion of the whole dataset.

For prompts, do not use the quality tags of Anima, but you can use tag prompting or natural language.

After full training was run, the model seems more competent on variety of sampler + scheduler combinations, but still falls apart in more complex scenes. I would say it is a fairly competent 1girl generator for the amount of effort put in though. Whatever was left of text producing capacity is mostly nuked out and I think the model defaults to photographic generation now.