Skip to main content
Text Stage

Text Stage · screenshot coming soon

​
What this node does

Text Stage is ComfyTV’s text generator. Write what you want the LLM to do in main_prompt, pick a workflow (a ComfyUI subgraph behind the scenes), click ▶ Run, and the generated string appears in the node preview. The output type is COMFYTV_TEXT—not a native ComfyUI string socket, but a project snapshot you can wire to texts on Image Stage or Video Stage so it merges into their prompts. Text Stage can also take upstream texts, images, and videos as multimodal context (whether they are used depends on the workflow and preset bindings).

​
When to use it

  • Expand a short idea with an LLM, then feed Image / Video Stage for t2i or t2v.
  • Generate product copy, narration, character bios, or other plain text.
  • Merge multiple upstream text outputs via texts, appended to main_prompt with newlines.
  • Draft story premises (Storyboard Stage has its own main_prompt too).

​
How ComfyTV stages work

  • A stage with ▶ Run executes only that node—not the whole ComfyUI graph via global Queue.
  • After a run, the text snapshot is stored in the project; downstream stages read it on their next run and do not automatically re-run Text Stage.
  • The workflow dropdown lists subgraphs from workflows/text/. ComfyTV maps main_prompt, seed, max_length, etc. into that subgraph.
Text workflows have no SaveImage; ComfyTV reads the LLM node’s first string output via graph_output_first in the preset.

​
Types (COMFYTV_TEXT vs native ComfyUI)

Conversion:
  • Native → ComfyTV: ComfyTV/Bridge → → ComfyTV Text (Run to snapshot)
  • ComfyTV → native: ← ComfyTV Text (load snapshot for native nodes)
See bridges.md.

​
Parameters

​
workflow

Backend from workflows/text/. Built-in: Local Qwen3 4B (Qwen3 4B chat via TextGenerate). Empty or grey list → restart ComfyUI or check workflow files. Models: models.md.

​
main_prompt

Your main instruction. Upstream texts append with newlines. Example: Write three short scene descriptions: a café at dawn, a rainy street, neon at night.

​
texts / images / videos

Autogrow slots for upstream stage outputs. Text is most common; images/videos are for multimodal workflows (Local Qwen3 4B is primarily text).

​
custom_params

Extra parameters (max_length, seed, temperature) configured in the ComfyTV sidebar when the stage is selected—mapped into the subgraph.

​
project_id / parent_output_id / force_run_token

UI-managed; bind Project context and output lineage.

​
Outputs

​
Step by step for beginners

  1. Add Project, then ComfyTV → Generate → Text Stage.
  2. workflow → Local Qwen3 4B (or your installed backend).
  3. main_prompt: Write a 20-word cyberpunk opening narration.
  4. Ensure Qwen3 4B is downloaded (models.md).
  5. Click ▶ Run and wait for text in the preview.
  6. Add Image Stage; wire text → texts (or paste into main_prompt).
  7. Run Image Stage to generate from the LLM output.
This page covers one node only. For end-to-end workflows, multi-stage pipelines, type conversion, and design rationale, see the ComfyTV user guides on GitHub:

​
Repository and workflows

​
FAQ

Q: Run finished but no text / “no runner registered”?
A: Workflow label mismatch. Restart ComfyUI or re-bind the preset in the sidebar. Q: Cannot connect text to Image Stage?
A: Use texts or a COMFYTV_TEXT-compatible socket—not native STRING. Use Bridge if needed. Q: Changed main_prompt but Image still uses old text?
A: Snapshots—downstream does not re-run Text Stage. Re-run Text Stage, then Image Stage.
  • Image Stage / Video Stage—consume COMFYTV_TEXT as prompt context.
  • Storyboard Stage—LLM output as structured storyboard JSON, not plain text.
  • Bridge → / ← ComfyTV Text—interop with native ComfyUI text.
  • Project—project context.