dverdu-freepik commited on
Commit
04b7e5f
·
1 Parent(s): a389f19

fix: Update README

Browse files
Files changed (1) hide show
  1. README.md +10 -7
README.md CHANGED
@@ -14,25 +14,28 @@ tags:
14
  ![Flux.1 Lite](sample_images/flux1-lite-8B_sample.png)
15
 
16
  # Flux.1 Lite
 
17
 
18
- We are thrilled to announce the alpha release of Flux.1 Lite, an 8B parameter transformer model distilled from the FLUX.1-dev model. This version uses 7 GB less RAM and runs 23% faster while maintaining the same precision (bfloat16) as the original model.
 
 
 
19
 
20
  ![Flux.1 Lite vs FLUX.1-dev](sample_images/models_comparison.png)
21
 
22
  ## Text-to-Image
23
 
24
- Flux.1 Lite is ready to unleash your creativity! For the best results, we strongly **recommend using a `guidance_scale` of 3.5 and setting `n_steps` between 22 and 30**.
25
 
26
  ```python
27
  import torch
28
  from diffusers import FluxPipeline
29
 
30
- base_model_id = "Freepik/flux.1-lite-8B-alpha"
31
  torch_dtype = torch.bfloat16
32
  device = "cuda"
33
 
34
  # Load the pipe
35
- model_id = "Freepik/flux.1-lite-8B-alpha"
36
  pipe = FluxPipeline.from_pretrained(
37
  model_id, torch_dtype=torch_dtype
38
  ).to(device)
@@ -40,7 +43,7 @@ pipe = FluxPipeline.from_pretrained(
40
  # Inference
41
  prompt = "A close-up image of a green alien with fluorescent skin in the middle of a dark purple forest"
42
 
43
- guidance_scale = 3.5 # Keep guidance_scale at 3.5
44
  n_steps = 28
45
  seed = 11
46
 
@@ -78,14 +81,14 @@ Stay tuned! Our goal is to distill FLUX.1-dev further until it can run smoothly
78
  We've also crafted a ComfyUI workflow to make using Flux.1 Lite even more seamless! Find it in `comfy/flux.1-lite_workflow.json`.
79
  ![ComfyUI workflow](comfy/flux.1-lite_workflow.png)
80
 
81
- The safetensors checkpoint is available here: [flux.1-lite-8B-alpha.safetensors](flux.1-lite-8B-alpha.safetensors)
82
 
83
  ## Try it out at Freepik!
84
 
85
  Our [AI generator](https://www.freepik.com/pikaso/ai-image-generator) is now powered by Flux.1 Lite!
86
 
87
  ## 🔥 News 🔥
88
-
89
  * Oct 23, 2024. Alpha 8B checkpoint is publicly available on [HuggingFace Repo](https://huggingface.co/Freepik/flux.1-lite-8B-alpha).
90
 
91
  ## Citation
 
14
  ![Flux.1 Lite](sample_images/flux1-lite-8B_sample.png)
15
 
16
  # Flux.1 Lite
17
+ We are thrilled to announce the release of Flux.1 Lite, an 8B parameter transformer model distilled from the FLUX.1-dev model. This version uses 7 GB less RAM and runs 23% faster while maintaining the same precision (bfloat16) as the original model.
18
 
19
+ `🔥 UPDATE 🔥`: We have released a new version of Flux.1 Lite 8B. This version is trained with a new dataset and achieves better results than the previous alpha version. The main changes include:
20
+ - Distillation for a broader range of guidance values (2.0-5.0)
21
+ - Distillation for a broader range of number of steps (20-32)
22
+ - More diverse dataset with longer prompts
23
 
24
  ![Flux.1 Lite vs FLUX.1-dev](sample_images/models_comparison.png)
25
 
26
  ## Text-to-Image
27
 
28
+ Flux.1 Lite is ready to unleash your creativity! For the best results, we strongly **recommend using a `guidance_scale` between 2.0 and 5.0 and setting `n_steps` between 20 and 32**.
29
 
30
  ```python
31
  import torch
32
  from diffusers import FluxPipeline
33
 
 
34
  torch_dtype = torch.bfloat16
35
  device = "cuda"
36
 
37
  # Load the pipe
38
+ model_id = "Freepik/flux.1-lite-8B"
39
  pipe = FluxPipeline.from_pretrained(
40
  model_id, torch_dtype=torch_dtype
41
  ).to(device)
 
43
  # Inference
44
  prompt = "A close-up image of a green alien with fluorescent skin in the middle of a dark purple forest"
45
 
46
+ guidance_scale = 3.5
47
  n_steps = 28
48
  seed = 11
49
 
 
81
  We've also crafted a ComfyUI workflow to make using Flux.1 Lite even more seamless! Find it in `comfy/flux.1-lite_workflow.json`.
82
  ![ComfyUI workflow](comfy/flux.1-lite_workflow.png)
83
 
84
+ The safetensors checkpoint is available here: [flux.1-lite-8B.safetensors](flux.1-lite-8B.safetensors)
85
 
86
  ## Try it out at Freepik!
87
 
88
  Our [AI generator](https://www.freepik.com/pikaso/ai-image-generator) is now powered by Flux.1 Lite!
89
 
90
  ## 🔥 News 🔥
91
+ * Dec 30, 2024. Flux.1 Lite 8B new trained model is publicly available on [HuggingFace Repo](https://huggingface.co/Freepik/flux.1-lite-8B).
92
  * Oct 23, 2024. Alpha 8B checkpoint is publicly available on [HuggingFace Repo](https://huggingface.co/Freepik/flux.1-lite-8B-alpha).
93
 
94
  ## Citation