stability_sdk 0.2.11 → 0.2.12

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 431d7a70dc0a5a538abdbe9a486bf5c632d1e6a8e7e58dd7fedf1fa89b15722b
-  data.tar.gz: a6d5c05cf4915e7c91ceeb92864b2cc84f852ec40a61ce174ba57b3a97ec0df9
+  metadata.gz: d5aac33cb8d504aab26b515dc2fa6bd9d76c70829a0aa706eabc6be39043b143
+  data.tar.gz: ca6a3cd1cd24763a9f9427132a4e8fde4c688d9bc9d3409ef921139fa62fa232
 SHA512:
-  metadata.gz: 7c5ba5862fd6446e243cdf54760fe596a4298e2d2860183df4cef847cc75d4414e9ceb21e8bd927e71e0fbacca9ce742628107197ca1e7c380abbbabff1e7cdc
-  data.tar.gz: 69b3cef394cbd5bd784bfd8953b1bd8721e4921e96f3a7bda79c0de3a733692853005008e1946cad75b24889df5dc1c8f9479238baac488a37a64ef49dbe39b7
+  metadata.gz: 95b76f6caf8e5a32af792cc0d284f81b76c7e38c682b24a5fc91204b3373096a47a98ab1ee6f0e32cbbd2039ad6b57c920c262661cda221c54eac529bd54e51a
+  data.tar.gz: f196c8b879b36e69c84d40c2528fd9fbf7f07e1ae6eec9c4a996a66a8a73841ea50d73d8ca941b7813ca9d5fbada94140143619bdb841de4ce9f71316af1303e

data/README.md

@@ -61,9 +61,25 @@ Options:
     -m, --mask_image=VAL             path to mask image
         --start_schedule=VAL         start schedule for init image (must be greater than 0, 1 is full strength text prompt, no trace of image). default 1.0
         --end_schedule=VAL           end schedule for init image. default 0.01
+        --guidance_preset=VAL        Guidance preset to use. See generation.GuidancePreset for supported values. default `GUIDANCE_PRESET_NONE`
+        --guidance_cuts=VAL          Number of cuts to use for guidance. default 0
+        --guidance_strength=VAL      Strength of the guidance. We recommend values in range [0.0,1.0]. A good default is 0.25. default nil
+        --guidance_prompt=VAL        Prompt to use for guidance, defaults to `YOUR_PROMPT_TEXT` argument (above) if not specified.
+        --guidance_models=VAL        Models to use for guidance. default nil
     -v, --verbose
 ```
 
+#### CLIP guidance
+
+By specifying `--guidance_preset` and other `--guidance_*` options, you can use CLIP guidance, which is enabled by default on DreamStudio's web interface.
+See also [this notebook](https://github.com/Stability-AI/stability-sdk/blob/c04381f960008f37c7392467cfaabfdf8f763e6a/nbs/demo_colab.ipynb) for more details.
+
+```sh
+# example
+# [note] --steps=35 is recommended. which is also the default value of the web interface
+STABILITY_SDK_API_KEY=xxx stability-client --guidance_preset=GUIDANCE_PRESET_FAST_BLUE --guidance_strength=0.25 --steps=35 --sampler=k_dpm_2_ancestral "A dream of a distant galaxy, by Caspar David Friedrich, matte painting trending on artstation HQ"
+```
+
 ### SDK usage
 
 This sample code saves a generated image as `result.png`.
@@ -116,6 +132,32 @@ After checking out the repo, run `bin/setup` to install dependencies. Then, run
 
 To install this gem onto your local machine, run `bundle exec rake install`. To release a new version, update the version number in `version.rb`, and then run `bundle exec rake release`, which will create a git tag for the version, push git commits and tags, and push the `.gem` file to [rubygems.org](https://rubygems.org).
 
+### update submodule to try a new feature on Stability-AI/api-interfaces
+
+- update submodule
+
+```sh
+git submodule update --init  # if you haven't fetched the content of the submodule yet
+cd api-interfaces
+
+# checkout some branch/commit you need
+git fetch
+git reset --hard origin/some_branch
+
+cd ..
+```
+
+- build
+
+```sh
+bundle exec rake protoc
+
+git diff
+# now you may be able to confirm that the diff is created in lib/generation_pb.rb
+```
+
+- modify the `lib/stability_sdk/client.rb` to try some new features
+
 ## Contributing
 
 Bug reports and pull requests are welcome on GitHub at https://github.com/cou929/stability-sdk-ruby.

data/exe/stability-client

@@ -18,20 +18,25 @@ opt.banner = "Usage: stability-client [options] YOUR_PROMPT_TEXT"
 opt.separator ""
 opt.separator "Options:"
 opt.on("--api_key=VAL", "api key of DreamStudio account. You can also specify by a STABILITY_SDK_API_KEY environment variable") {|v| options[:api_key] = v }
-opt.on("-H", "--height=VAL", "height of image in pixel. default 512") {|v| options[:height] = v }
-opt.on("-W", "--width=VAL", "width of image in pixel. default 512") {|v| options[:width] = v }
-opt.on("-C", "--cfg_scale=VAL", "CFG scale factor. default 7.0") {|v| options[:cfg_scale] = v }
-opt.on("-A", "--sampler=VAL", "ddim, plms, k_euler, k_euler_ancestral, k_heun, k_dpm_2, k_dpm_2_ancestral, k_lms. default k_lms") {|v| options[:sampler] = v }
-opt.on("-s", "--steps=VAL", "number of steps. default 50") {|v| options[:steps] = v }
-opt.on("-S", "--seed=VAL", "random seed to use in integer") {|v| options[:seed] = v }
-opt.on("-p", "--prefix=VAL", "output prefixes for artifacts. default `generation`") {|v| options[:prefix] = v }
+opt.on("-H", "--height=VAL", Integer, "height of image in pixel. default 512") {|v| options[:height] = v }
+opt.on("-W", "--width=VAL", Integer, "width of image in pixel. default 512") {|v| options[:width] = v }
+opt.on("-C", "--cfg_scale=VAL", Float, "CFG scale factor. default 7.0") {|v| options[:cfg_scale] = v }
+opt.on("-A", "--sampler=VAL", String, "ddim, plms, k_euler, k_euler_ancestral, k_heun, k_dpm_2, k_dpm_2_ancestral, k_lms. default k_lms") {|v| options[:sampler] = v }
+opt.on("-s", "--steps=VAL", Integer, "number of steps. default 50") {|v| options[:steps] = v }
+opt.on("-S", "--seed=VAL", Integer, "random seed to use in integer") {|v| options[:seed] = v }
+opt.on("-p", "--prefix=VAL", String, "output prefixes for artifacts. default `generation`") {|v| options[:prefix] = v }
 opt.on("--no-store", "do not write out artifacts") {|v| options[:no_store] = v }
-opt.on("-n", "--num_samples=VAL", "number of samples to generate") {|v| options[:num_samples] = v }
-opt.on("-e", "--engine=VAL", "engine to use for inference. default `stable-diffusion-v1`") {|v| options[:engine_id] = v }
-opt.on("-i", "--init_image=VAL", "path to init image") {|v| options[:init_image] = v }
-opt.on("-m", "--mask_image=VAL", "path to mask image") {|v| options[:mask_image] = v }
-opt.on("--start_schedule=VAL", "start schedule for init image (must be greater than 0, 1 is full strength text prompt, no trace of image). default 1.0") {|v| options[:start_schedule] = v }
-opt.on("--end_schedule=VAL", "end schedule for init image. default 0.01") {|v| options[:end_schedule] = v }
+opt.on("-n", "--num_samples=VAL", Integer, "number of samples to generate. default 1") {|v| options[:num_samples] = v }
+opt.on("-e", "--engine=VAL", String, "engine to use for inference. default `stable-diffusion-v1-5`") {|v| options[:engine_id] = v }
+opt.on("-i", "--init_image=VAL", String, "path to init image") {|v| options[:init_image] = v }
+opt.on("-m", "--mask_image=VAL", String, "path to mask image") {|v| options[:mask_image] = v }
+opt.on("--start_schedule=VAL", Float, "start schedule for init image (must be greater than 0, 1 is full strength text prompt, no trace of image). default 1.0") {|v| options[:start_schedule] = v }
+opt.on("--end_schedule=VAL", Float, "end schedule for init image. default 0.01") {|v| options[:end_schedule] = v }
+opt.on("--guidance_preset=VAL", String,"Guidance preset to use. See generation.GuidancePreset for supported values. default `GUIDANCE_PRESET_NONE`") {|v| options[:guidance_preset] = v }
+opt.on("--guidance_cuts=VAL", Integer, "Number of cuts to use for guidance. default 0") {|v| options[:guidance_cuts] = v }
+opt.on("--guidance_strength=VAL", Float, "Strength of the guidance. We recommend values in range [0.0,1.0]. A good default is 0.25. default nil") {|v| options[:guidance_strength] = v }
+opt.on("--guidance_prompt=VAL", String, "Prompt to use for guidance, defaults to `YOUR_PROMPT_TEXT` argument (above) if not specified.") {|v| options[:guidance_prompt] = v }
+opt.on("--guidance_models=VAL", Array, "Models to use for guidance. default nil") {|v| options[:guidance_models] = v }
 opt.on("-v", "--verbose") { logger.level = Logger::DEBUG }
 opt.parse!(ARGV)
 
@@ -48,7 +53,7 @@ if options.has_key?(:init_image)
   raise StabilitySDK::InvalidParameter, "width and height of init_image must be a multiple of 64" if size[0] % 64 != 0 || size[1] % 64 != 0
 end
 
-client = StabilitySDK::Client.new(api_key: options[:api_key])
+client = StabilitySDK::Client.new(api_key: options[:api_key], logger: logger)
 
 client.generate(prompt, options) do |answer|
   StabilitySDK::CLI.save_answer(answer, options, logger)

data/lib/stability_sdk/cli.rb

@@ -23,10 +23,15 @@ module StabilitySDK
           logger.warn "not implemented for ArtifactType #{artifact.type}"
         end
 
+        if artifact.finish_reason == :FILTER
+          logger.debug "the generated image is filtered"
+        end
+
         next if filename == "" || contents == ""
 
         File.open(filename, "wb") do |f|
           f.write(contents)
+          logger.debug "wrote #{artifact.type} to #{filename}"
         end
       end
     end

data/lib/stability_sdk/client.rb

@@ -1,5 +1,6 @@
 require "grpc"
 require "generation_services_pb"
+require "logger"
 
 module StabilitySDK
   class Client
@@ -36,29 +37,37 @@ module StabilitySDK
         stub_params[kw] = options[kw] if options.has_key?(kw)
       end
 
+      if options.has_key?(:logger)
+        @logger = options[:logger]
+      else
+        logger = Logger.new(STDOUT)
+        logger.level = Logger::WARN
+        @logger = logger
+      end
+
       @stub = Gooseai::GenerationService::Stub.new(host, creds, **stub_params)
     end
 
     def generate(prompt, options, &block)
-      width = options.has_key?(:width) ? options[:width].to_i : DEFAULT_IMAGE_WIDTH
-      height = options.has_key?(:height) ? options[:height].to_i : DEFAULT_IMAGE_HEIGHT
+      width = options.has_key?(:width) ? options[:width] : DEFAULT_IMAGE_WIDTH
+      height = options.has_key?(:height) ? options[:height] : DEFAULT_IMAGE_HEIGHT
 
       if width % 64 != 0 || height % 64 != 0
         raise InvalidParameter, "width and height must be a multiple of 64"
       end
 
-      samples = options.has_key?(:num_samples) ? options[:num_samples].to_i : DEFAULT_SAMPLE_SIZE
-      steps = options.has_key?(:steps) ? options[:steps].to_i : DEFAULT_STEPS
-      seed = options.has_key?(:seed) ? [options[:seed].to_i] : [rand(4294967295)]
+      samples = options.has_key?(:num_samples) ? options[:num_samples] : DEFAULT_SAMPLE_SIZE
+      steps = options.has_key?(:steps) ? options[:steps] : DEFAULT_STEPS
+      seed = options.has_key?(:seed) ? [options[:seed]] : [rand(4294967295)]
       transform = Gooseai::TransformType.new(
-        diffusion: options.has_key?(:sampler) ? SAMPLER_ALGORITHMS[options[:sampler]] : DEFAULT_SAMPLER_ALGORITHM,
+        diffusion: options.has_key?(:sampler) ? SAMPLER_ALGORITHMS[options[:sampler].to_sym] : DEFAULT_SAMPLER_ALGORITHM,
       )
-      parameters = [Gooseai::StepParameter.new(
+      step_parameter = Gooseai::StepParameter.new(
         scaled_step: 0,
         sampler: Gooseai::SamplerParameters.new(
-          cfg_scale: options.has_key?(:cfg_scale) ? options[:cfg_scale].to_f : DEFAULT_CFG_SCALE,
+          cfg_scale: options.has_key?(:cfg_scale) ? options[:cfg_scale] : DEFAULT_CFG_SCALE,
         ),
-      )]
+      )
 
       prompt_param = []
       if prompt != ""
@@ -66,21 +75,65 @@ module StabilitySDK
       end
       if options.has_key?(:init_image)
         prompt_param << init_image_to_prompt(options[:init_image])
-        parameters = [Gooseai::StepParameter.new(
-          scaled_step: 0,
-          sampler: Gooseai::SamplerParameters.new(
-            cfg_scale: options.has_key?(:cfg_scale) ? options[:cfg_scale].to_f : DEFAULT_CFG_SCALE,
-          ),
-          schedule: Gooseai::ScheduleParameters.new(
-            start: options.has_key?(:start_schedule) ? options[:start_schedule].to_f : DEFAULT_START_SCHEDULE,
-            end: options.has_key?(:end_schedule) ? options[:end_schedule].to_f : DEFAULT_END_SCHEDULE,
-          ),
-        )]
+        step_parameter.scaled_step = 0
+        step_parameter.sampler = Gooseai::SamplerParameters.new(
+          cfg_scale: options.has_key?(:cfg_scale) ? options[:cfg_scale] : DEFAULT_CFG_SCALE,
+        )
+        step_parameter.schedule = Gooseai::ScheduleParameters.new(
+          start: options.has_key?(:start_schedule) ? options[:start_schedule] : DEFAULT_START_SCHEDULE,
+          end: options.has_key?(:end_schedule) ? options[:end_schedule] : DEFAULT_END_SCHEDULE,
+        )
       end
       if options.has_key?(:mask_image)
         prompt_param << mask_image_to_prompt(options[:mask_image])
       end
 
+      # CLIP guidance
+      if options.has_key?(:guidance_preset) && options[:guidance_preset] != "GUIDANCE_PRESET_NONE"
+        step_parameter.sampler = nil
+
+        guidance_prompt =
+          if options.has_key?(:guidance_prompt)
+            Gooseai::Prompt.new(text: options[:guidance_prompt])
+          else
+            Gooseai::Prompt.new(text: prompt)
+          end
+
+        guidance_strength = nil
+        if options.has_key?(:guidance_strength) && options[:guidance_strength] != 0
+          guidance_strength = options[:guidance_strength]
+        end
+
+        models = nil
+        if options.has_key?(:guidance_models)
+          models = options[:guidance_models].map { |m| Gooseai::Model.new(alias: m) }
+        end
+
+        cutouts = nil
+        if options.has_key?(:guidance_cuts)
+          cutouts = Gooseai::CutoutParameters.new(count: options[:guidance_cuts])
+        end
+
+        step_parameter.guidance = Gooseai::GuidanceParameters.new(
+          guidance_preset: Gooseai::GuidancePreset.const_get(options[:guidance_preset].to_sym),
+          instances: [
+            Gooseai::GuidanceInstanceParameters.new(
+              guidance_strength: guidance_strength,
+              models: models,
+              cutouts: cutouts,
+              prompt: guidance_prompt,
+            ),
+          ],
+        )
+
+        if transform.diffusion != Gooseai::DiffusionSampler::SAMPLER_K_DPM_2_ANCESTRAL && transform.diffusion != Gooseai::DiffusionSampler::SAMPLER_K_EULER_ANCESTRAL
+          transform = Gooseai::TransformType.new(
+            diffusion: Gooseai::DiffusionSampler::SAMPLER_K_DPM_2_ANCESTRAL
+          )
+          @logger.warn "CLIP guidance is only supported with ancestral samplers. So override it with SAMPLER_K_DPM_2_ANCESTRAL."
+        end
+      end
+
       image_param = Gooseai::ImageParameters.new(
         width: width,
         height: height,
@@ -88,7 +141,7 @@ module StabilitySDK
         steps: steps,
         seed: seed,
         transform: transform,
-        parameters: parameters,
+        parameters: [step_parameter],
       )
 
       req = Gooseai::Request.new(
@@ -97,8 +150,20 @@ module StabilitySDK
         image: image_param
       )
 
+      @logger.debug "sending request."
+      start = Time.now
       @stub.generate(req).each do |answer|
+        duration = Time.now - start
+        if answer.artifacts.size > 0
+          artifact_types = answer.artifacts.map { |a| a.type }
+          @logger.debug "got #{answer.answer_id} with #{artifact_types} in #{duration.round(2)}s"
+        else
+          @logger.debug "got keepalive #{answer.answer_id} in #{duration.round(2)}s"
+        end
+
         block.call(answer)
+
+        start = Time.now
       end
     end
 

data/lib/stability_sdk/version.rb

@@ -1,3 +1,3 @@
 module StabilitySDK
-  VERSION = "0.2.11"
+  VERSION = "0.2.12"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: stability_sdk
 version: !ruby/object:Gem::Version
-  version: 0.2.11
+  version: 0.2.12
 platform: ruby
 authors:
 - Kosei Moriyama
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2022-10-28 00:00:00.000000000 Z
+date: 2022-10-31 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: grpc