MOSS BBOX Prompter

MOSS BBOX Prompter

English | 日本語 · v2.0.1 · Release <img width="1789" height="741" alt="{EBFA18EE-0E04-416C-A085-46EEBC310FFD}" src="https://github.com/user-attachments/assets/a9736fd8-202e-40e0-b12a-c83cd5ba8fd1" />

Arrange colored boxes in ComfyUI and write a prompt for each character, object or text area. Canvas defines placement and Prompter supplies words; connect both to the same Export to build the MOSS-Image prompt.

Seven colors, multiple boxes of the same color, Angle/Framing/Focus, dialogue styles, reference images and Undo/Redo are included. V2 adds reference-image analysis → automatic box text → editing after OFF, while retaining V1 workflows and saved settings.

Install and open a workflow

1. In ComfyUI-Manager → Install via Git URL, enter this repository.
2. Save your work, finish running jobs, restart ComfyUI normally, then press Ctrl+F5. Use a ComfyUI version with MOSS-Image support.
3. Load a workflow below and select your installed model, TE, VAE and LoRA files. The V2 template needs rgthree-comfy and Core Generate Text / If/Else Switch / Seed nodes. No separate Krea2 installation is required.

| Workflow | Download | Use | | --- | --- | --- | | V2 · 2.0.1 | ZIP with reference image / JSON | Current template. Analysis starts OFF; ordinary prompting works as before. | | V1 · 1.0.11 | Previous release / Template JSON | Previous setup. Older release history is retained. |

Extract the V2 ZIP and open examples/MOSS_BBOX_Canvas_V2.json. Upload/select the included ed6c61ad22695e639fb4b5b75fd2b931.webp in Load Image, or use your own reference. New nodes start with blank prompts; sample workflow wording is not copied into them.

Reference-image analysis

1. Choose an image and a box

The generation subgraph exposes Image analysis enable, target and kind, alongside the usual generation settings.

| target — destination box | kind — analysis subject | | --- | --- | | | | | Auto uses the first created/stored box and displays its color, such as Auto · RED. Select a box explicitly to keep the destination fixed. | character describes a person; object describes an object. This does not change the card's Type. |

Missing box means there is no valid target. With analysis ON, running raises an error: create a box or select an existing one first. This version analyzes one reference image and one box. Text targets are excluded, including when Auto selects a Text box.

2. Turn ON and Run

Set Image analysis enable = true, then press the ordinary Run button. The analyzed text is saved automatically to the selected card and used for image generation. No copy/paste or separate Apply button is needed.

3. Turn OFF to edit

Set Image analysis enable = false. The text stays in the card, becomes Manual, and can be edited or replaced. Run again to generate from that saved text, as with ordinary prompting.

Running again with analysis ON overwrites manual edits with the new analysis. Switching ON alone does not analyze anything. Keep it OFF while using your edited wording; save the whole workflow to retain it.

Safety in v2.0.1: changing the image, target, body, coordinates, Type or Switch invalidates pending results; Undo or OFF→ON does not revive them. An in-flight job still finishes with its queued inputs. Update from GitHub, restart ComfyUI and press Ctrl+F5.

<details> <summary>ON / OFF control screenshots</summary>

ON (true): Run analyzes the image and saves the result to the selected target.

OFF (false): Run uses the saved card text; it does not reanalyze the image.

</details>

The reference is described as text; it is not direct image conditioning. Exact identity, lettering or reproduction is not guaranteed. Even while OFF, connected Load Image inputs must contain a valid image. Analysis settings and detailed behavior.

Everyday controls

| Task | Operation | | --- | --- | | Edit a box | Use its card. Type selects Character / Object / Text; Framing, Angle and Focus add composition hints. Text keeps unused controls disabled. | | Select a covered box | Use the Canvas right-click menu or Alt+click to cycle overlapping boxes. The menu also provides foreground ordering. | | Move / copy / protect | Multi-select, align/distribute, Alt-drag to duplicate, and Lock. Canvas Undo/Redo restores edits. | | Dialogue | Use a separate Text box for exact words. Set Usage/Shape; open Lettering / Bubbles for colors, direction and style. 😀 inserts emoji. Use Export text_mode=literal. | | Shared settings / name | Normally only actual box cards appear. Open Shared defaults when needed; card ⋯ contains Name and Shared. Shared replacement previews the change and asks for confirmation. | | Save | Save both Canvas and Prompter in the workflow. Per-box text, Type, styles, expansion and manual editor sizes are retained. |

Optional connections

• External STRING: text-processing/LLM node → MOSS BBOX Text Link.text; Canvas.framing_data → Text Link.framing_data; select Box; Text Link.external_text → Prompter.external_text. The card previews external text read-only. Unlinking restores its saved manual text. Details.
• Krea2 effects: Krea2 BBOX Prompter Suite optionally adds Background Effect / BBOX Prompt Effect after Export. Connect Export.prompt_text → Effect.prompt_in → Effect.prompt_out → CLIP Text Encode.text. Chain both effects if desired; use prompt_out to keep the original prompt. Wiring guide.

Detailed guide and older-version notes · Validation · Issues

V2 GPU operation was confirmed by the user. CPU/UI, saved-state and packaging checks are recorded separately in the validation report. GitHub Releases and Registry publication are separate.

MIT License · UI adapted from Krea2 BBOX Prompter Suite. See LICENSE and NOTICE.md.