theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-03-22 08:40:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

support flux highresfix

Browse Source
This commit is contained in:
Artiprocher
2024-08-19 13:35:27 +08:00
parent 80aa4d8e19
commit 778a2d8f84
7 changed files with 111 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/image_synthesis/README.md
View File

@@ -2,10 +2,20 @@
Image synthesis is the base feature of DiffSynth Studio. We can generate images with very high resolution.
### Example: FLUX
Example script: [`flux_text_to_image.py`](./flux_text_to_image.py)
|1024*1024 (original)|1024*1024 (classifier-free guidance)|2048*2048 (highres-fix)|
|-|-|-|
|![image_1024](https://github.com/user-attachments/assets/d8e66872-8739-43e4-8c2b-eda9daba0450)|![image_1024_cfg](https://github.com/user-attachments/assets/1073c70d-018f-47e4-9342-bc580b4c7c59)|![image_2048_highres](https://github.com/user-attachments/assets/8719c1a8-b341-48c1-a085-364c3a7d25f0)|
### Example: Stable Diffusion
Example script: [`sd_text_to_image.py`](./sd_text_to_image.py)
LoRA Training: [`../train/stable_diffusion/`](../train/stable_diffusion/)
|512*512|1024*1024|2048*2048|4096*4096|
|-|-|-|-|
|![512](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/55f679e9-7445-4605-9315-302e93d11370)|![1024](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/6fc84611-8da6-4a1f-8fee-9a34eba3b4a5)|![2048](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/9087a73c-9164-4c58-b2a0-effc694143fb)|![4096](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/edee9e71-fc39-4d1c-9ca9-fa52002c67ac)|
@@ -14,6 +24,8 @@ Example script: [`sd_text_to_image.py`](./sd_text_to_image.py)
Example script: [`sdxl_text_to_image.py`](./sdxl_text_to_image.py)
LoRA Training: [`../train/stable_diffusion_xl/`](../train/stable_diffusion_xl/)
|1024*1024|2048*2048|
|-|-|
|![1024](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/67687748-e738-438c-aee5-96096f09ac90)|![2048](https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/584186bc-9855-4140-878e-99541f9a757f)|

25
examples/image_synthesis/flux_text_to_image.py
View File

@@ -12,9 +12,30 @@ model_manager.load_models([
])
pipe = FluxImagePipeline.from_model_manager(model_manager)
prompt = "A captivating fantasy magic woman portrait set in the deep sea. The woman, with blue spaghetti strap silk dress, swims in the sea. Her flowing silver hair shimmers with every color of the rainbow and cascades down, merging with the floating flora around her. Smooth, delicate and fair skin."
negative_prompt = "worst quality, low quality, monochrome, zombie, interlocked fingers, Aissist, dim, fuzzy, depth of Field, nsfw,"
# Disable classifier-free guidance (consistent with the original implementation of FLUX.1)
torch.manual_seed(6)
image = pipe(
"A captivating fantasy magic woman portrait set in the deep sea. The woman, with blue spaghetti strap silk dress, swims in the sea. Her flowing silver hair shimmers with every color of the rainbow and cascades down, merging with the floating flora around her. Smooth, delicate and fair skin.",
num_inference_steps=30
prompt=prompt,
num_inference_steps=30,
)
image.save("image_1024.jpg")
# Enable classifier-free guidance
torch.manual_seed(6)
image = pipe(
prompt=prompt, negative_prompt=negative_prompt,
num_inference_steps=30, cfg_scale=2.0
)
image.save("image_1024_cfg.jpg")
# Highres-fix
torch.manual_seed(7)
image = pipe(
prompt=prompt,
num_inference_steps=30,
input_image=image.resize((2048, 2048)), height=2048, width=2048, denoising_strength=0.6, tiled=True
)
image.save("image_2048_highres.jpg")