ideogram

name

ideogram

description

Ideogram image nodes in Nebula — v4 generate, masked inpaint (ideogram-edit), maskless edit-prompt, remix, reframe, replace-background, character, upscale, transparent, remove-background, describe, magic-prompt. Activate when the user configures any ideogram-* node, asks about Ideogram in Nebula, or wants to wire or test mask-painter → ideogram-edit inpainting on the canvas. Covers dual-route auth (IDEOGRAM_API_KEY direct vs FAL_KEY fallback), mask polarity (BLACK = edit), the four-wire canvas graph, Inspector mask painting, expand_prompt, preset image execution, auto-wire of base image, and how to verify a visible inpaint result. Sourced from docs/api-guides/ideogram.md and live canvas E2E on feat/inpainting (2026-06-10).

Ideogram in Nebula

When to use

• User configures or asks about any ideogram-* node.
• User wants masked inpainting on the canvas (mask-painter → ideogram-edit).
• User says the edit "looks the same" / no visible output / mask node shows a weird B&W square.
• Agent is wiring, smoke-testing, or demoing the inpaint flow for the user.

API param details and route tables: docs/api-guides/ideogram.md. FAL dialect notes: .claude/skills/fal/SKILL.md.

Universal rules

1. Dual route. IDEOGRAM_API_KEY → direct api.ideogram.ai; else FAL_KEY → FAL slugs. Inspector shows the param set for the active route.
2. Mask polarity: BLACK = edit (both routes). On mask-painter, set Painted Area Means → Black = edit (Ideogram). Painted strokes are stored white-on-black in _maskData; the engine inverts at export for Ideogram.
3. Mask dimensions must match the base image exactly. Mask Painter resizes to upstream image size — always wire the same base image into ideogram-edit.image and into mask-painter.image.
4. Output is on the edit node. After Run, ideogram-edit shows the result inside its card (state: complete, outputs.image). No downstream output node required.

Canvas inpaint — agent playbook

Read references/canvas-inpaint.md before wiring or running inpaint for a user. It is the step-by-step checklist agents kept getting wrong.

Minimum graph (four wires)

text-input ──text──► ideogram-edit.prompt
image-input ──image──► mask-painter.image
image-input ──image──► ideogram-edit.image    ← REQUIRED (mask-painter does NOT pass the photo)
mask-painter ──mask──► ideogram-edit.mask

Connecting mask → edit auto-wires upstream image → edit.image when that port is empty (frontend graphStore.onConnect). Still verify all four edges before Run.

Mask painting (Inspector only)

1. Select mask-painter → Inspector → Paint Mask (not on the canvas node body).
2. Paint a filled region over everything you want changed — not a hairline stroke on empty background.
3. Bad: thin diagonal on white table + prompt "gold coin" → invisible even when the run succeeds.
4. Good: brush over the label / object / face you want replaced; use a thick brush or fill the area.

Params before Run

Node	Set
mask-painter	polarity: black-edit
ideogram-edit	expand_prompt: true (or magic_prompt: AUTO on direct route) for stronger interpretation
text-input	Prompt describes what should appear in the masked region

Image input

• Uploaded images: need filePath (set automatically on upload).
• Preset thumbnails: _previewUrl like /api/presets/thumbnails/studio-product is enough — execute normalizes to filePath on the backend.

Run + verify

1. Toolbar Run (or executeGraph / POST /api/execute).
2. All nodes should reach complete. If mask-painter errors with "Image input is required", the base image wire is missing or image-input has no resolvable file.
3. Visible inpaint check: output should differ inside the masked bbox, not just globally. If the masked region looks unchanged, the mask is too small, on the wrong surface, or unfilled — have the user repaint (or repaint yourself via Inspector), don't rerun with the same mask.
4. ideogram-edit output is often a different resolution (e.g. 1024² from 640²) but composition should show a clear change in the painted area when done right.

When masked inpaint is the wrong tool

• Replace label text / remove object without drawing a mask → ideogram-edit-prompt (maskless, prompt-driven edit).
• FLUX-style fill (white = edit) → flux-fill-inpaint + mask-painter polarity White = edit.

Node quick map

Node	Use
ideogram-v4	Text-to-image, typography-first
ideogram-edit	Masked inpaint — prompt + image + mask
ideogram-edit-prompt	Maskless edit ("remove the lamp post")
ideogram-remix	Image + prompt restyle (v4 direct)
ideogram-reframe	Outpaint to new aspect (direct needs resolution enum)
ideogram-replace-background	Subject + new background prompt
ideogram-character	One character_reference_images per request
ideogram-upscale	Upscale with optional prompt

Live smoke (backend)

cd backend && .venv/bin/python scripts/smoke_ideogram_live.py
# dual-route remainder only:
.venv/bin/python scripts/smoke_ideogram_live.py --remaining-dual-route

Requires IDEOGRAM_API_KEY and/or FAL_KEY in Settings.

Agent mistakes (do not repeat)

Mistake	Symptom	Fix
Only image → mask → edit chain	Weak/global edit or validation failure	Also wire image → edit.image (or re-connect mask to trigger auto-wire)
Thin stroke on empty white	Run succeeds, looks identical	Filled mask over the target object/text
Raw B&W mask on mask-painter card	Confusing preview	Expected before composite fix; node should show source + overlay when wired
_previewUrl only, no Run	No output on edit node	Run graph; output appears on ideogram-edit when complete
expand_prompt: false + vague prompt	Subtle/no visible change	Enable expand / stronger prompt
Programmatic mask = 1px line	API succeeds, invisible edit	Use filled ellipse/rectangle over target region in test scripts