Thought Bubble:
ThoughtBubbleNode turns text arranged on a canvas into prompts for an image-generation workflow. It selects a canvas box, processes its text into positive and negative prompts, and can prepare those prompts for image generation. It also supports LoRA requests and persistent text changes identified during parsing, while reusing previously processed results when the relevant model, CLIP, LoRA configuration, and prompts have not changed. This lets you work with prompt text on the canvas while keeping the resulting prompts and generation components together.
Thought Bubble Input Parameters:
canvas_data
Required STRING containing the canvas state as JSON. The node reads its boxes and chooses the first maximized box; if none is maximized, it chooses a box titled output; otherwise, it uses the first box with nonblank content, or the first box if all are blank. It then parses the chosen box’s content. The default is a JSON canvas with one empty text box titled output, positioned at x 100 and y 100, sized 400 by 300, with normal display state; it also includes pan at x 0 and y 0, zoom 1.0, grid size 100, the grid shown, no saved view, and an empty theme. No minimum or maximum length is specified. Invalid JSON prevents prompt parsing, and an empty chosen box produces empty prompt text.
model
Optional MODEL connection. Supply a model when your workflow needs the node to pass one onward or apply LoRAs requested by the parsed prompt. Without a LoRA request, the supplied model passes through unchanged; when LoRAs are requested, the node may return a modified model. There is no configured default model, minimum, maximum, or list of model options. If you leave it disconnected, the model output starts as None.
clip
Optional CLIP connection. Supply CLIP to turn the resulting positive and negative text into conditioning that downstream generation nodes can use. The node may also apply prompt-requested LoRAs to the connected CLIP before encoding. There is no configured default CLIP, minimum, maximum, or list of CLIP options. If you leave it disconnected, conditioning remains an empty list unless a parsed LoRA request supplies a usable CLIP through the LoRA application process.
Thought Bubble Output Parameters:
model
MODEL result for the rest of your workflow. It is the incoming model when no LoRA is applied, or the model returned by the LoRA application process when the parsed prompt requests LoRAs. If you supplied no model, it may be None.
clip
CLIP result used by this node to encode the prompts and available for downstream connections. It is the incoming CLIP when no LoRA is applied, or the CLIP returned by the LoRA application process when LoRAs are requested. Without a usable CLIP, the node cannot generate conditioning.
positive
CONDITIONING generated by tokenizing and encoding the parsed positive prompt with the resulting CLIP. Connect it where your workflow expects positive conditioning. If no usable CLIP is available, this output is an empty list; when CLIP is available, even an empty positive prompt is sent to CLIP for encoding.
negative
CONDITIONING generated by tokenizing and encoding the parsed negative prompt with the resulting CLIP. Connect it where your workflow expects negative conditioning. If no usable CLIP is available, this output is an empty list; when CLIP is available, even an empty negative prompt is sent to CLIP for encoding.
positive_prompt_text
STRING containing the positive prompt text returned by the canvas parser. Use it to inspect or display the wording that the node passes to CLIP for positive conditioning. It is empty when no source content is parsed or parsing fails.
negative_prompt_text
STRING containing the negative prompt text returned by the canvas parser. Use it to inspect or display the wording that the node passes to CLIP for negative conditioning. It is empty when parsing produces no negative text, no source content is parsed, or parsing fails.
Thought Bubble Usage Tips:
- Put the text you want parsed in the box the node will select. A maximized box takes priority over a box titled
output; when neither applies, the first box with nonblank content is selected. - Connect both
modelandclipwhen you want a conventional model-and-conditioning path. In particular, connectclipwhen you need usablepositiveandnegativeconditioning. - Check
positive_prompt_textandnegative_prompt_textwhen results differ from what you expected. They show the parser’s returned wording separately from the encoded conditioning. - Keep
canvas_dataas valid JSON. The default canvas is valid but contains no prompt text, so add content to the selected box before expecting a nonempty prompt. - The node requests execution on every run so persistent values can change. Its caches can reuse matching LoRA and conditioning results within the node instance, but this does not mean the node itself is skipped.
Thought Bubble Common Errors and Solutions:
[Thought Bubble Error] Could not decode JSON data from canvas.
- Explanation: The
canvas_datastring is not valid JSON, so the node cannot read its boxes. The prompt text remains empty; conditioning can still be generated from empty text if CLIP is connected. - Solution: Restore valid canvas data or correct the JSON syntax, including quotation marks, commas, and brackets.
[Thought Bubble Error] Loading wildcards: <details>
- Explanation: The node could not create or read the
wildcardsfolder under the application’s user directory, or could not read one of its.txtfiles. The specific cause appears in place of<details>. - Solution: Check that the user directory is accessible and writable, and that wildcard text files can be opened as UTF-8. Use the reported details to identify the affected file or permission.
[Thought Bubble Error] <details>
- Explanation: An error other than invalid JSON occurred while processing the canvas, parsing its text, applying LoRAs, or encoding conditioning. The node prints the underlying exception in place of
<details>and returns whatever results were available at that point. - Solution: Read the full console message first. Check the selected box’s content and any referenced files or LoRAs; if the error concerns encoding, verify that a working CLIP is connected.
[Thought Bubble Warning] Could not broadcast persister update: <details>
- Explanation: The parser changed persistent prompt text, but the node could not send that update to the live canvas interface. Prompt processing may still have succeeded, while the visible canvas may not reflect the change.
- Solution: Check the full warning in the console and confirm whether the canvas updated. If it did not, preserve your prompt text and retry after checking the application’s server connection.
No error message; prompt text is empty
- Explanation: The chosen box may be empty. For example, a maximized empty box is selected before a populated
outputbox, and the default canvas starts with empty content. - Solution: Add text to the selected box, or change which box is maximized. Check the prompt-text outputs to confirm what was parsed.
No error message; conditioning is empty
- Explanation: When the resulting CLIP is
None, the node leaves both conditioning outputs as empty lists. - Solution: Connect a usable CLIP input and run the workflow again.
