Krea 2 Turbo @ 4 steps (w/ my 4-step LoRA) vs MOSS SFT at 4 steps (ww AK's 4-step LoRA) - same prompt set you've seen me publish in the past (headline: MOSS is really good! and great job by AK on his LoRA)
retained text
As many of you, I have also been keeping a close eye on the MOSS model released recently and have been quite impressed by its results. I have also been closely monitoring the 8 and 4 steps LoRAs for MOSS that have been published recently. I decided to do a full run of ***my prompts*** (that I was using during my Krea 2 Turbo 2 and 4 Steps LoRAs development) on MOSS with [AK's 4 step LoRA](https://huggingface.co/akhaliq/MOSS-Image-SFT-4-steps) and present to you the results in detail here: [https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA/blob/main/assets/comparison-with-moss/README.md](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA/blob/main/assets/comparison-with-moss/README.md), where you can see full size images by both models compared. # Models and LoRAs ||Krea 2 Turbo|MOSS-Image SFT| |:-|:-|:-| |**Model**|[krea/Krea-2-Turbo](https://huggingface.co/krea/Krea-2-Turbo): Krea's Turbo text-to-image model, CFG-free, 8 steps as shipped|[OpenMOSS-Team/MOSS-Image-SFT](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT): OpenMOSS's text-to-image model (5B DiT, Qwen3.5-9B text encoder), 50 steps with CFG 3 as shipped| |**4-step LoRA**|[this LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA): 4 steps, strength 1.0|[akhaliq/MOSS-Image-SFT-4-steps](https://huggingface.co/akhaliq/MOSS-Image-SFT-4-steps): 4 steps, strength 1.0| |**Here**|4 steps · no CFG · 4 NFE|4 steps · no CFG · 4 NFE| I personally think, MOSS model is really good, especially for its 5.08B size compared to Krea 2 Turbo's 12.82B. It just means you get a lot more, quicker and with less compute. What do you think? Share your renders with MOSS and Krea 2 Turbo and your thoughts on which one is better (***if we have to compare at all***, both models have their place in the open source space, if you ask me, I mean I should be biased even towards Krea considering I spent months training the LoRAs for it, but I really do like MOSS and what it gives us). My full comparison at HF: [https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA/blob/main/assets/comparison-with-moss/README.md](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA/blob/main/assets/comparison-with-moss/README.md)
document fetched 2026-10-20 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1x9xs4c/krea_2_turbo_4_steps_w_my_4step_lora_vs_moss_sft