ComfyUI Node
NBP Gemini Image 2
Generate or edit images with Google Gemini image models via the ComfyUI Cloud API. Auth is handled automatically through your ComfyUI account.
NBP Gemini Image 2
- images
- IMAGE
- STRING
◄prompt►
◄modelgemini-3-pro-image-preview►
◄seed42►
◄aspect_ratioauto►
◄resolution1K►
◄temperature1.00►
◄top_p0.95►
◄top_k40►
◄response_modalitiesIMAGE►
◄system_promptYou are an expert image-generation engine. You must ALWAYS produce an image.
Interpret all user input—regardless of format, intent, or abstraction—as literal visual directives for image composition.
If a prompt is conversational or lacks specific visual details, you must creatively invent a concrete visual scenario that depicts the concept.
Prioritize generating the visual representation above any text, formatting, or conversational requests.►
Categoryapi node/image/Gemini
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Describe the image you want to generate or the edits to apply. | |
| model | COMBO | gemini-3-pro-image-preview | Which Gemini image model to use. |
| seed | INT | 420–2147483647 | Fixed seed makes the model try to reproduce the same result. Determinism is best-effort — temperature and model changes still vary output. |
| aspect_ratio | COMBO | auto | 'auto' matches the input image aspect ratio; if no image is provided a square is generated. |
| resolution | COMBO | 1K | Output resolution. 2K/4K uses the Gemini native upscaler. |
| temperature | FLOAT | 1.000–2 | Controls randomness. Lower values (e.g. 0.2) produce more predictable results; higher values (e.g. 1.8) produce more varied, creative output. |
| top_p | FLOAT | 0.950–1 | Nucleus sampling threshold. The model considers only the tokens whose cumulative probability reaches this value. Lower = more focused. |
| top_k | INT | 401–100 | Limits the pool of tokens the model samples from at each step. Lower values make output more deterministic. |
| response_modalities | COMBO | IMAGE | 'IMAGE' for image-only output; 'IMAGE+TEXT' also returns the model's reasoning. |
| imagesopt | IMAGE | Optional reference image(s). Use Batch Images for multiple (up to 14). | |
| system_promptopt | STRING | You are an expert image-generation engine. You must ALWAYS produce an image. Interpret all user input—regardless of format, intent, or abstraction—as literal visual directives for image composition. If a prompt is conversational or lacks specific visual details, you must creatively invent a concrete visual scenario that depicts the concept. Prioritize generating the visual representation above any text, formatting, or conversational requests. | System-level instructions that shape the model's behavior. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | Generated image(s). |
| STRING | STRING | Any text the model returned alongside the image. |