cjwbw/supir

Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild. This version uses LLaVA-13b for captioning.

Input

Configure the inputs for the AI model.

Seed

Random seed. Leave blank to randomize the seed

Image *

Low quality input image.

S Cfg

Classifier-free guidance scale for prompts.

S Churn

100

Original churn hy-param of EDM.

S Noise

100

Original noise hy-param of EDM.

Upscale

100

Upsampling ratio of given inputs.

A Prompt

Additive positive prompt for the inputs.

Min Size

100

Minimum resolution of output images.

N Prompt

Negative prompt for the inputs.

S Stage1

100

Control Strength of Stage1 (negative means invalid).

S Stage2

100

Control Strength of Stage2.

Edm Steps

500

Number of steps for EDM Sampling Schedule.

Use Llava

Enable

Use LLaVA model to get captions.

Linear Cfg

Enable

Linearly (with sigma) increase CFG from 'spt_linear_CFG' to s_cfg.

model_name

Choose a model. SUPIR-v0Q is the default training settings with paper. SUPIR-v0F is high generalization and high image quality in most cases. Training with light degradation settings. Stage1 encoder of SUPIR-v0F remains more details when facing light degradations.

color_fix_type

Color Fixing Type..

Spt Linear Cfg

100

Start point of linearly increasing CFG.

Linear S Stage2

Enable

Linearly (with sigma) increase s_stage2 from 'spt_linear_s_stage2' to s_stage2.

Spt Linear S Stage2

100

Start point of linearly increasing s_stage2.

Output

The generated output will appear here.

No output yet

Click "Generate" to create an output.