Moss-Image SFT +new DMD2 lora (for SFT) , 8 Steps
retained text
base model + dmd2 for base gens here [https://www.reddit.com/r/StableDiffusion/comments/1xarnzg/mossimage\_base\_new\_dmd2\_lora\_8\_steps/](https://www.reddit.com/r/StableDiffusion/comments/1xarnzg/mossimage_base_new_dmd2_lora_8_steps/) all parameters for all pics were kept same as that post , only the models and lora were swapped. SFT gens are sharper and do not have the bokeh issue of base + dmd2 base gens. However , they also feel a bit style collapsed. for example 1. the cinematic still (pic7) doesn't evoke the texture of the 80s analog film in the prompt which is there in original post. 2. the last pic does not feel like a 90s DnD artstyle , in original it looks much closer.
document fetched 2026-10-21 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1xb5fqe/mossimage_sft_new_dmd2_lora_for_sft_8_steps