import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("nebulette/mossrael-2b-encoder", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]The direct continuation of Moqs for the Moss:RAEL finetune:
- JSON layout support (without the text field)
- 400 art genres and techniques
- reference names as either "style of name", "art by @name"
Backward compatible with MOSSad.
The layout editor has been used to define the pose in the second row of images.
History
This is the fourth release based on the Qwen 2B model, now with improved prompt weighting, better support for:
- single word tags,
- common character names (written in either capital or lowercase letters),
- danbooru bigrams,
- English bigrams,
- @name tags in a comma-separated and natural language context.
Source data
- common colors, verbs
- hundreds of names
- danbooru_tags_classified
- danbooru_tag_taxonomy
- danbooru2026_moe
- danbooru_wiki_2606
- gelbooru 40k (each sample has a new, unexplored booru tag)
- openart_painterly_foundations
- pexels_portrait
- saber (2610.22602)
- sdxl_artists
- spright_coco
- z_image_ethnicity
- Downloads last month
- 9
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for nebulette/mossrael-2b-encoder
Base model
OpenMOSS-Team/MOSS-Image-Base