> ## Documentation Index
> Fetch the complete documentation index at: https://docs.comfy.org/llms.txt
> Use this file to discover all available pages before exploring further.

# SenseNovaConditioning - ComfyUI Built-in Node Documentation

> Prepare SenseNova conditioning for image generation or for interleaved text-and-image output.

Prepare SenseNova conditioning for image generation or for interleaved text-and-image output. The `task` selector sets the prompt template and the stopping behavior, and the node reports whether the model wants an image, so a loop knows when to run the sampler.

## Inputs

| Parameter | Description | Data Type | Required | Range |
| - | - | - | - | - |
| `clip` | SenseNova CLIP model. Connect the CLIP loader that also feeds CLIP Text Encode; the encoder output goes to `conditioning`. Required for every task except when the incoming conditioning is empty or already finished. | CLIP | No | - |
| `conditioning` | SenseNova text conditioning from CLIP Text Encode. | CONDITIONING | Yes | - |
| `negative` | Optional negative conditioning. It is updated and returned for the `interleave` and `self-correction` tasks. | CONDITIONING | No | - |
| `task` | Task to run. `"image"` encodes the prompt for a standard sampler, `"image thinking"` lets the model think before it answers, `"interleave"` produces text and images in one pass, and `"self-correction"` lets the model fix its own previous output. | DYNAMIC\_COMBO | Yes | `"image"` `"image thinking"` `"interleave"` `"self-correction"` |
| `max_tokens` | Maximum tokens generated during this call. Default: 1024. | INT | Yes | 1 to 8192 |

### `interleave` Inputs

These inputs appear when `task` is set to `"interleave"`.

| Parameter | Description | Data Type | Required | Range |
| - | - | - | - | - |
| `thinking` | Let the model think before it answers. Default: True. | BOOLEAN | Yes | True False |
| `samples` | Previous sampler output, connected when continuing after an image request. | LATENT | No | - |

### `self-correction` Inputs

These inputs appear when `task` is set to `"self-correction"`.

| Parameter | Description | Data Type | Required | Range |
| - | - | - | - | - |
| `samples` | Previous sampler output, connected when continuing after an image request. | LATENT | No | - |

**Constraints and behavior:**

* The `clip` input takes the SenseNova CLIP model itself, not the CLIP Text Encode output, which is `CONDITIONING`. The node returns early without using `clip` when the incoming conditioning is empty or already finished; otherwise a missing `clip`, or one that is not a SenseNova CLIP, is an error.
* `"image thinking"` needs exactly one positive conditioning entry; `"interleave"` and `"self-correction"` need one positive and one negative entry.
* When the model is waiting for an image, connect the preceding sampler's LATENT output to `samples`. Passing samples before an image was requested is an error.
* When the incoming conditioning is already finished, the node returns it unchanged with an empty `generated_text` and `image_requested` set to False.

## Outputs

| Output Name | Description | Data Type |
| - | - | - |
| `conditioning` | Conditioning for the sampler, extended with the text generated by this call. | CONDITIONING |
| `generated_text` | Text generated by this call. Use SenseNova Interleave Decode for the complete result. | STRING |
| `negative` | Updated negative conditioning for the `interleave` and `self-correction` tasks. | CONDITIONING |
| `image_requested` | True when the model asked for an image: run the sampler, then feed the loop back into this node. | BOOLEAN |

> This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! [Edit on GitHub](https://github.com/Comfy-Org/embedded-docs/blob/main/comfyui_embedded_docs/docs/SenseNovaConditioning/en.md)

***

**Source fingerprint (SHA-256):** `97fcdc26ccf80f48c5628f695a4ae8231b065b67414c93dec9bced6bb8d4c77c`


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.