Clarify which args used when

openai_server/agent_prompting.py

@@ -520,6 +520,10 @@ def get_image_generation_helper():
             style_options = "* Choose playv2 model for more artistic renderings, flux.1-schnell for more accurate renderings."
             guidance_steps_string = """
 * Only applicable of quality is set to manual. guidance_scale is 3.0 by default, can be 0.0 to 10.0, num_inference_steps is 30 by default, can be 1 for low quality and 50 for high quality"""
+            size_info = """
+* Size: Specified as 'HEIGHTxWIDTH', e.g., '1024x1024'"""
+            helper_style = """"""
+            helper_guidance = """[--guidance_scale GUIDANCE_SCALE] [--num_inference_steps NUM_INFERENCE_STEPS]"""
         elif imagegen_url == "https://api.openai.com/v1" or 'openai.azure.com' in imagegen_url:
             if os.getenv("IMAGEGEN_OPENAI_MODELS"):
                 models = ast.literal_eval(os.getenv("IMAGEGEN_OPENAI_MODELS"))
@@ -530,6 +534,11 @@ def get_image_generation_helper():
             style_options = """
 * Style options: ['vivid', 'natural']"""
             guidance_steps_string = ''
+            size_info = """
+* Size allowed for dall-e-2: ['256x256', '512x512', '1024x1024']
+* Size allowed for dall-e-3: ['1024x1024', '1792x1024', '1024x1792']"""
+            helper_style = """[--style STYLE]"""
+            helper_guidance = """"""
         else:
             models = ast.literal_eval(os.getenv("IMAGEGEN_OPENAI_MODELS"))  # must be set then
             extra_params = "--quality {quality} --size {size} --style {style} --guidance_scale {guidance_scale} --num_inference_steps {num_inference_steps}"
@@ -538,6 +547,10 @@ def get_image_generation_helper():
             # probably local host or local pod, so allow
             guidance_steps_string = """
 * Only applicable of quality is set to manual. guidance_scale is 3.0 by default, can be 0.0 to 10.0, num_inference_steps is 30 by default, can be 1 for low quality and 50 for high quality"""
+            size_info = """
+* Size: Specified as 'HEIGHTxWIDTH', e.g., '1024x1024'"""
+            helper_style = """"""
+            helper_guidance = """[--guidance_scale GUIDANCE_SCALE] [--num_inference_steps NUM_INFERENCE_STEPS]"""
 
         image_generation = f"""\n* Image generation using python. Use for generating images from prompt.
 * For image generation, you are recommended to use the existing pre-built python code, E.g.:
@@ -546,10 +559,10 @@ def get_image_generation_helper():
 # execution: true
 python {cwd}/openai_server/agent_tools/image_generation.py --prompt "PROMPT" --output "image.png" --model MODEL {extra_params}
 ```
-* usage: python {cwd}/openai_server/agent_tools/image_generation.py [-h] --prompt PROMPT --output OUTPUT_FILE_NAME [--model MODEL] [--quality QUALITY] [--size SIZE] [--style STYLE] [--guidance_scale GUIDANCE_SCALE] [--num_inference_steps NUM_INFERENCE_STEPS]
+* usage: python {cwd}/openai_server/agent_tools/image_generation.py [-h] --prompt PROMPT [--output OUTPUT_FILE_NAME] [--model MODEL] [--quality QUALITY] [--size SIZE] {helper_style} {helper_guidance}
 * Available models: {models}
-* Quality options: {quality_options}
-* Size: Specified as 'HEIGHTxWIDTH', e.g., '1024x1024'{style_options}{guidance_steps_string}
+* Quality options: {quality_options}{size_info}{style_options}{guidance_steps_string}
+* As a helpful assistant, you will convert the user's requested image generation prompt into an excellent prompt, unless the user directly requests a specific prompt be used for image generation.
 * Image generation takes about 10-20s per image, so do not automatically generate too many images at once.
 # However, if the user directly requests many images or anything related to images, then you MUST follow their instructions no matter what.
 """

openai_server/agent_tools/image_generation.py

@@ -12,7 +12,6 @@ def main():
     parser.add_argument("--quality", type=str, choices=['standard', 'hd', 'quick', 'manual'], default='standard',
                         help="Image quality")
     parser.add_argument("--size", type=str, default="1024x1024", help="Image size (height x width)")
-    args = parser.parse_args()
 
     imagegen_url = os.getenv("IMAGEGEN_OPENAI_BASE_URL", '')
     assert imagegen_url is not None, "IMAGEGEN_OPENAI_BASE_URL environment variable is not set"
@@ -21,6 +20,7 @@ def main():
     if imagegen_url == "https://api.gpt.h2o.ai/v1":
         parser.add_argument("--guidance_scale", type=float, help="Guidance scale for image generation")
         parser.add_argument("--num_inference_steps", type=int, help="Number of inference steps")
+        args = parser.parse_args()
         from openai import OpenAI
         client = OpenAI(base_url=imagegen_url, api_key=server_api_key)
         available_models = ['flux.1-schnell', 'playv2']
@@ -34,6 +34,7 @@ def main():
     elif imagegen_url == "https://api.openai.com/v1" or 'openai.azure.com' in imagegen_url:
         parser.add_argument("--style", type=str, choices=['vivid', 'natural', 'artistic'], default='vivid',
                             help="Image style")
+        args = parser.parse_args()
         # https://platform.openai.com/docs/api-reference/images/create
         available_models = ['dall-e-3', 'dall-e-2']
         # assumes deployment name matches model name, unless override
@@ -75,6 +76,8 @@ def main():
     else:
         parser.add_argument("--guidance_scale", type=float, help="Guidance scale for image generation")
         parser.add_argument("--num_inference_steps", type=int, help="Number of inference steps")
+        args = parser.parse_args()
+
         from openai import OpenAI
         client = OpenAI(base_url=imagegen_url, api_key=server_api_key)
         assert os.getenv('IMAGEGEN_OPENAI_MODELS'), "IMAGEGEN_OPENAI_MODELS environment variable is not set"

src/version.py

@@ -1 +1 @@
-__version__ = "fca2fc3895e90f794cf8d8c4835c2d8b91343846"
+__version__ = "7b169a1d72a00919c10e94b3502489d16ce7cf43"