Skip to content

Latest commit

 

History

History
194 lines (160 loc) · 8.28 KB

README.md

File metadata and controls

194 lines (160 loc) · 8.28 KB
Raw

Fooocus-API-XYZ: Variable Parameter Image Generator for Fooocus-API

Generates a series of Fooocus images with unrestricted user supplied parameter variations.

Requires Fooocus-API installed and running.

Quick Start

Get it:

git clone https://github.com/daedon/Fooocus-API-XYZ

Directory Fooocus-API-XYZ will be created, rename it to whatever you like.

Run it:

cd Fooocus-API-XYZ
python3 xyz.py myJob

The generated images will be saved in the Fooocus-API output directory.

Creating a new job

To create a new job, simply run xyz with the new job name:

cd Fooocus-API-XYZ
python3 xyz.py newJobName

The first time you use a jobName:

  • A directory with the job's name is created in "Fooocus-API-XYZ/jobs".
  • Associated sub-directories will be created.
  • Default files will be copied to "jobs/jobName" (from the default_ directories).
  • Add, Modify image generation parameters in "jobs/jobName/parms".
  • Customize the curl template "jobs/jobName/curl.template" (each job has it's own template).
  • To set default parameters for all new jobs created, edit the files in the default_ directories.

Example with 3 variable parameters

Parameter files are located in the ../Fooocus-API-XYZ/jobs/myJob/parms directory. The table below contains 5 parameters from the demo job. The first 3 characters of the parameter file name serve to sort and determine the order the parameters are processed. In the example below, resolution, guidance_scale and sharpness will automatically be used to build save_name since they vary (have more than one value). "steps" has only one value and won't vary but will be included in save_name since it ends with an "_". All permutations of the following parameters will be used to make curl calls to Fooocus-API.

File -> _0_steps_ _1_resolution _2_guidance_scale _3_sharpness ___image_number
line 1 25 512*640 4.0 2.0 1
line 2 640*640 4.5 3.0
line 3 1024*1024

The parameters in the table above will produce the following save_names:

myJob_0001_steps_25_resolution_512x512_guidance_scale_4.0_sharpness_2.0_<time stamp>
myJob_0002_steps_25_resolution_512x512_guidance_scale_4.0_sharpness_3.0_<time stamp>
myJob_0003_steps_25_resolution_512x512_guidance_scale_4.5_sharpness_2.0_<time stamp>
myJob_0004_steps_25_resolution_512x512_guidance_scale_4.5_sharpness_3.0_<time stamp>
myJob_0005_steps_25_resolution_640x640_guidance_scale_4.0_sharpness_2.0_<time stamp>
myJob_0006_steps_25_resolution_640x640_guidance_scale_4.0_sharpness_3.0_<time stamp>
myJob_0007_steps_25_resolution_640x640_guidance_scale_4.5_sharpness_2.0_<time stamp>
myJob_0008_steps_25_resolution_640x640_guidance_scale_4.5_sharpness_3.0_<time stamp>
myJob_0009_steps_25_resolution_1024x1024_guidance_scale_4.0_sharpness_2.0_<time stamp>
myJob_0010_steps_25_resolution_1024x1024_guidance_scale_4.0_sharpness_3.0_<time stamp>
myJob_0011_steps_25_resolution_1024x1024_guidance_scale_4.5_sharpness_2.0_<time stamp>
myJob_0012_steps_25_resolution_1024x1024_guidance_scale_4.5_sharpness_3.0_<time stamp>

save_name can be configured with saveNameCFG in config.py. When saveNameCFG is 1 or 2, parameters with more than 1 value will automatically be included in save_name. To force the inclusion of a parameter with only one value, terminate the parameter's file name with an "_", eg ___seed_.

saveNameCFG = 0: save_name contains jobName, image number and time stamp. 
saveNameCFG = 1: save_name contains also the variable parameter value.
saveNameCFG = 2: save_name contains both the variable parameter name and value.

0: myJob_0001_024844
1: myJob_0001_25_512x512_4.0_2.0_<time stamp>
2: myJob_0001_steps_25_resolution_512x512_guidance_scale_4.0_sharpness_2.0_<time stamp>

Separators for save_name can be modified in config.py. Note the separator ---- between the parameters and the = between the paramater names and their value.

myJob_0001----steps=25----resolution=512x512----guidance_scale=4.0----sharpness=2.0----<time stamp>

Summary:

  • Fooocus-API-XYZ requires that Fooocus-API be running.
  • Fooocus-API-XYZ is data driven.
  • Reads parameter values from parameter files in the job's parms subdirectory.
  • Each parameter file represents a Fooocus parameter: eg, ___steps, ___resolution, ___prompt, ___base_model.
  • The 1st character of each parameter file name must be an underscore "_".
  • The 1st 3 characters of each parameter file name are used for documentation and sorting purposes only.
  • The 1st 3 characters of each parameter file are discarded.
  • Parameter files are read in alphabetical order, hence the first 3 characters decide the order of the parameters.
  • Empty lines in Parameter files are ignored.
  • Lines beginning with a "#" Parameter files are comments and ignored.
  • No parameter values are read after a line with a "." in parameter files.
  • Each parameter file line represents 1 value for that parameter.
  • Parameters are read line by line into lists matching the parameter names.
  • Permutations of all the parameter values are generated.
  • The curl template has substitution parameters matching the paramater file names.
  • The curl template can be customized for each individual job (see Customzing below).
  • For each permutation, fills the curl template with the current iteration's values (___parameter).
  • The current iteration of curl.template is written to runCURL.
  • runCURL is made executable and executed.

Configuring

Directory locations and other parameters can be changed in config.py.

If image_number is set to N, Fooocus will generate N images per curl call with sequential seed numbers. To reduce repetiion in large permuations and create N images with random seeds, insert N lines in ___seed each with a -1.

Customizing

The curl template has close to 100 parameters, only a dozen or so are set up by default.

For example, sampler_name is hard coded in the curl job template.

"sampler_name": "dpmpp_2m_sde_gpu",

To make sampler_name one of your variable parameters in myJob,

  • edit myJob's curl.template and replace dpmpp_2m_sde_gpu with ___sampler_name (keep the quotes).
"sampler_name": "___sampler_name",
  • In the myJob/parms directory, create a file named ___sampler_name with your sampler_name values.
  • Generate your images.
  • To have sampler_name as a varable parameter in all future jobs, make the above changes to the defaults directories.
graph TD;
    Read_Parameters[Go]-->fileA[Read Parameter File A];
    Read_Parameters-->fileB[Read Parameter File B];
    Read_Parameters-->fileC[Read Parameter File C];

    fileA-->listA[List A];
    fileB-->listB[List B];
    fileC-->listC[List C];

    listA-->A1[Parameter 1];
    listA-->A2[Parameter 2];
    listB-->B1[Parameter 1];
    listB-->B2[Parameter 2];
    listC-->C1[Parameter 1];
    listC-->C2[Parameter-2];


    A1-->Generate_Permutation[Generate Permutations];
    A2-->Generate_Permutation;
    B1-->Generate_Permutation;
    B2-->Generate_Permutation;
    C1-->Generate_Permutation;
    C2-->Generate_Permutation;

    Generate_Permutation-->A1_B1_C1[A1 B1 C1];
    Generate_Permutation-->A1_B1_C2[A1 B1 C2];
    Generate_Permutation-->A1_B2_C1[A1 B2 C1];
    Generate_Permutation-->A1_B2_C2[A1 B2 C2];
    Generate_Permutation-->A2_B1_C1[A2 B1 C1];
    Generate_Permutation-->A2_B1_C2[A2 B1 C2];
    Generate_Permutation-->A2_B2_C1[A2 B2 C1];
    Generate_Permutation-->A2_B2_C2[A2 B2 C2];

    A1_B1_C1-->template[Fill Curl Template];
    A1_B1_C2-->template;
    A1_B2_C1-->template;
    A1_B2_C2-->template;
    A2_B1_C1-->template;
    A2_B1_C2-->template;
    A2_B2_C1-->template;
    A2_B2_C2-->template;

    template-->CurlCall[execute runCURL];

    CurlCall-->ImageA1_B1_C1[Image 1];
    CurlCall-->ImageA1_B1_C2[Image 2];
    CurlCall-->ImageA1_B2_C1[Image 3];
    CurlCall-->ImageA1_B2_C2[Image 4];
    CurlCall-->ImageA2_B1_C1[Image 5];
    CurlCall-->ImageA2_B1_C2[Image 6];
    CurlCall-->ImageA2_B2_C1[Image 7];
    CurlCall-->ImageA2_B2_C2[Image 8];
Loading