nodes/silveroxides-ComfyUI_Gemini_Expanded_API
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat: Add bypass_mode to SSL_GeminiTextPrompt

Browse Source 
This commit introduces a new `bypass_mode` multiple choice input to the `SSL_GeminiTextPrompt` node. This feature allows you to choose whether to apply a special underscore padding to the "system_instruction", the "prompt", or "both" before sending them to the Gemini API. The padding format is `_c_h_a_r_a_c_t_e_r_s_`.

The following changes were made:
- Added `bypass_mode` to the `INPUT_TYPES` with options: "None", "system_instruction", "prompt", "both".
- Implemented a helper method `_pad_text_with_underscores` to perform the character padding.
- Modified the `generate` method to:
    - Accept `bypass_mode` as a parameter.
    - Conditionally apply padding to `system_instruction` and `prompt` based on the selected `bypass_mode`.
    - Use the (potentially) padded text in the API call to Gemini.
This commit is contained in:
google-labs-jules[bot]
2025-06-11 09:00:12 +00:00
parent 30926703e2
commit 80d5a8f2a9
1 changed files with 19 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gemini_nodes.py
+19 -4
View File
@@ -73,6 +73,7 @@ class SSL_GeminiTextPrompt(ComfyNodeABC):
"top_k": (IO.INT, {"default": 40, "min": 1, "max": 100, "step": 1}),
"max_output_tokens": (IO.INT, {"default": 8192, "min": 1, "max": 8192, "step": 1}),
"include_images": (IO.BOOLEAN, {"default": True}),
"bypass_mode": (["None", "system_instruction", "prompt", "both"], {"default": "None"}),
},
"optional": {
"input_image": (IO.IMAGE,),
@@ -90,6 +91,11 @@ class SSL_GeminiTextPrompt(ComfyNodeABC):
FUNCTION = "generate"
CATEGORY = "API/Gemini"
def _pad_text_with_underscores(self, text: str) -> str:
if not text:
return ""
return "_" + "_".join(list(text)) + "_"
def save_binary_file(self, data, mime_type):
ext = ".bin"
if mime_type == "image/png":
@@ -113,7 +119,7 @@ class SSL_GeminiTextPrompt(ComfyNodeABC):
tensor = torch.from_numpy(empty_image).unsqueeze(0)
return tensor
def generate(self, config, prompt, system_instruction, model, temperature, top_p, top_k, max_output_tokens, include_images,
def generate(self, config, prompt, system_instruction, model, temperature, top_p, top_k, max_output_tokens, include_images, bypass_mode,
input_image=None, input_image_2=None, use_proxy=False, proxy_host="127.0.0.1", proxy_port=7890,
use_seed=True, seed=0):
original_http_proxy = os.environ.get('HTTP_PROXY')
@@ -123,6 +129,15 @@ class SSL_GeminiTextPrompt(ComfyNodeABC):
print(f"[INFO] Starting generation, model: {model}, temperature: {temperature}")
# Pad text based on bypass_mode
padded_prompt = prompt
padded_system_instruction = system_instruction
if bypass_mode == "prompt" or bypass_mode == "both":
padded_prompt = self._pad_text_with_underscores(prompt)
if bypass_mode == "system_instruction" or bypass_mode == "both":
padded_system_instruction = self._pad_text_with_underscores(system_instruction)
actual_seed = None
if use_seed == True:
if seed == 0:
@@ -267,12 +282,12 @@ class SSL_GeminiTextPrompt(ComfyNodeABC):
img_part = {"inline_data": {"mime_type": "image/png", "data": img_bytes}}
img_parts.append(img_part)
contents = img_parts + [{"text": prompt}]
contents = img_parts + [{"text": padded_prompt}]
except Exception as e:
print(f"[ERROR] Error processing input image: {str(e)}")
return (f"Error processing input image: {str(e)}", self.generate_empty_image(), actual_seed if actual_seed is not None else 0)
else:
contents = prompt
contents = padded_prompt
response_modalities = ["text"]
if include_images == True:
@@ -303,7 +318,7 @@ class SSL_GeminiTextPrompt(ComfyNodeABC):
],
response_modalities=response_modalities,
system_instruction=[
types.Part.from_text(text=system_instruction),
types.Part.from_text(text=padded_system_instruction),
],
)