sd-inf commited on
Commit
64c5b4e
·
verified ·
1 Parent(s): f2e7b0e

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +66 -0
README.md ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ datasets:
4
+ - LucidityAI/meisho-sft
5
+ tags:
6
+ - image-generation
7
+ ---
8
+ # Meisho-S1-Preview
9
+
10
+ A Meisho S1 Preview is a generalist image-generation model based on [Tongyi-MAI/Z-Image](https://huggingface.co/Tongyi-MAI/Z-Image), finetuned on our [LucidityAI/meisho-sft dataset](https://huggingface.co/LucidityAI/meisho-sft)
11
+
12
+ Compared to it's base model, Meisho-S1-Preview handles areas like typography better, while having regressions in other places (see limitations), which we aim to resolve and surpass the base-model at in a later version, whether it be in a second version of a preview or the final S1 version, among the possibility of splitting Meisho into flash and pro versions, like our Synth 2.5 family.
13
+
14
+
15
+ ## Usage
16
+
17
+ You may use Meisho S1 Preview like so:
18
+
19
+ ```python
20
+ import torch
21
+ from diffsynth.pipelines.z_image import ZImagePipeline, ModelConfig
22
+
23
+ pipe = ZImagePipeline.from_pretrained(
24
+ torch_dtype=torch.bfloat16,
25
+ device="cuda",
26
+ model_configs=[
27
+ ModelConfig(path=["<repo>/transformer/diffusion_pytorch_model-00001-of-00002.safetensors",
28
+ "<repo>/transformer/diffusion_pytorch_model-00002-of-00002.safetensors"]),
29
+ ModelConfig(path=["<repo>/text_encoder/model-0000{}-of-00003.safetensors".format(i) for i in (1,2,3)]),
30
+ ModelConfig(path="<repo>/vae/diffusion_pytorch_model.safetensors"),
31
+ ],
32
+ tokenizer_config=ModelConfig(path="LucidityAI/Meisho-S1-Preview/tokenizer"),
33
+ )
34
+
35
+ image = pipe(
36
+ prompt="...",
37
+ seed=1234, rand_device="cuda",
38
+ num_inference_steps=50, cfg_scale=4.0,
39
+ height=1024, width=1024,
40
+ )
41
+ ```
42
+
43
+ Recommended settings: `num_inference_steps=50`, `cfg_scale=4.0`, 1024×1024. This is a
44
+ non-distilled base model, so it needs the full step count, unlike Z-Image-Turbo.
45
+
46
+ ## Examples
47
+
48
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_05_character_sheet.jpg">
49
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_10_type_poster.jpg">
50
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_12_food_flatlay.jpg">
51
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_11_color_palette.jpg">
52
+
53
+ ## Limitations
54
+
55
+ The preview version of Meisho-S1 struggles with:
56
+
57
+ - Detail
58
+ - Clothing
59
+ - Photorealistic people
60
+
61
+ ## Examples of Limitations
62
+
63
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_02_cafe_girl.jpg">
64
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_14_sprinter_bw.jpg">
65
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_04_magic_study.jpg">
66
+ <img style="height: 350px;" src="https://huggingface.co/LucidityAI/Meisho-S1-Preview/resolve/main/samples/compare_15_knit_macro.jpg">