Amazing Great job Devs
Using
-The provided comfy workflow for MOSS Image
-moss_image_sft_int8_convrot,
-qwen3.5_9b_int8_convrot and
-flux2-vae.safetensors
I'm really loving this MOSS Image SFT setup. I tried the qwen_image_2.1_int8_convrot, but It feels so commercial and pre made. I tried Flux2 also, same thing: Not as good as MOSS.
With MOSS I feel like I'm able to express my ideas with art work because it follows the idea I'm trying to express. Maybe this has to do with the text that the developers entered into the system prompt, or the model itself. What ever it is, they did a darn great job. I'm still experimenting with qwen_image_2.1, but so far I'm not feeling like it's giving me the same freedom as Moss Image. Maybe the generate text prompt that Qwen uses is doing too much thinking on it's own. But it's not just that, I love the images, the people MOSS produces. To the developers: Please don't change this. Again, I don't know if the system prompt has something to do with my success or the model itself, but I feel connected, like the training of the MOSS model is some how able to read my mind. I don't feel that way with any of these other ones, Z-Image, juggernautXL_ragnarok. They're great in their own way, but MOSS Image just blew me away in the last two days.
I've got the full 16 versions backed up also, but I need to upgrade my GPU/vram