Views
No views yet
0.99995, and MiniT2I-L/16 uses EMA decay 0.9999; both are exported from step 290K. Load one repository, then select the model at inference time with model_type.model_type | Model | Directory |
|---|---|---|
b16 | MiniT2I-B/16 | minit2i-b-16/ |
l16 | MiniT2I-L/16 | minit2i-l-16/ |
b, base, minit2i-b/16, l, large, and minit2i-l/16 are also supported.1import torch
2from diffusers import DiffusionPipeline
3
4HUB_MODEL_ID = "MiniT2I/MiniT2I"
5
6pipe = DiffusionPipeline.from_pretrained(
7 HUB_MODEL_ID,
8 custom_pipeline=HUB_MODEL_ID,
9 trust_remote_code=True,
10)
11
12image = pipe(
13 "A lonely astronaut standing on a quiet beach under two moons.",
14 model_type="b16",
15 guidance_scale=2.5,
16 num_inference_steps=100,
17 torch_dtype=torch.bfloat16,
18).images[0]
19image.save("minit2i-b16.png")
20
21image = pipe(
22 "a watercolor painting of a mountain lake at sunrise",
23 model_type="l16",
24 guidance_scale=6.0,
25 num_inference_steps=100,
26 torch_dtype=torch.bfloat16,
27).images[0]
28image.save("minit2i-l16.png")model_type="b16" does not download the L/16 weights.MiniT2I/MiniT2I-B-16-jax for MiniT2I-B/16MiniT2I/MiniT2I-L-16-jax for MiniT2I-L/161@misc{minit2i2026,
2 title = {MiniT2I: A Minimalist Baseline for Text-to-Image Synthesis},
3 author = {Wang, Xianbang and Zhao, Hanhong and Lu, Yiyang and Zhou, Kangyang and Ma, Linrui and He, Kaiming},
4 year = {2026},
5 url = {https://peppaking8.github.io/#/post/minit2i}
6}