Update: This 2.0 might be the last DogeMix, in fact this model should hold a while until I prepare the next realistic model which will be a whole different animal and won't be posted under "DogeMix". This is still a nice update over the 1.0 in every aspect. Euler A should be your bread and butter and if you're having trouble with some models, just use CFG rescale/ Dynamic Thresholding. Everything else, it should follow your prompt to the letter and blend nicely with your LoRAs.

Checkpoint focused on lora precision and characterization (age, ethnicity, clothing, body types, hairstyles). Pretty supportive of fantastic themes and also NSFW, well along the lines of the other two models.

Intuitive and responsive prompt solving, don't litter with useless words and things you don't want on the screen or you'll compromise the coherence of the generation. Image can be finetuned with minimal negatives like "worst quality/bad quality", use heavy embeddings at your own risk. Should give you no trouble on img2img and inpainting, but there will likely be an inpainting version in the future.

Base model I've been working for a while now, same stuff you can see at my pages or the civit samples (all the lora showcases are made on these 3 checkpoints, no cherrypicking, it goes as it is).

Recommended samplers:

Restart/DPM++ 2S a Karras around ~25 steps (excellent precision but sometimes grainy image)

DPM++ 2M Karras/DPM++ 3M SDE Karras around ~35 steps (sharper upscaling, perfect for photorealism but sometimes unresponsive/unreliable)

UniPC at 40+ steps (use it for intricate patterns, fine details at skin pores scale, same problem downside as the other K samplers)

Recommended upscalers:

4x_NMKD-Superscale-SP_178000_G for general purposes

R-ESRGAN General WDN 4xV3 preferably at 512 tiling for better details

Might open an article later for more details on prompting and sampling on these models, but until then you might have them figured out, pretty simple.