🕸️ Design Flow

● Unsaved ← Design 🖼️ Single image

❓ Design Flow — what each card does

The idea: place your real pictures where you want them (fast, no AI), then let a final step finish the image.
Images → 🟧 Compose (place layers) → 🟩 Final Render (clean · place · shadow). Tick ✨ AI finish to complete cut-offs (legs) + blend — identity-safe. Split the render into transparent layers with ✂️ Split Layer cards. Change wording UPSTREAM, before Compose.

🖼 Content — what appears in the image

🖼 Image — a source photo; mark it Subject or Background (for a Subject set place ◀L/■C/▶R + size). Kept pixel-exact. Example: drop Pete's photo → Subject → place ■ C.

🔤 TEXT — your words, with a mode button on the card. 🎨 Render graphic = AI renders them as a styled transparent layer for Compose. 📝 To prompt = doesn't render; the words + style are injected into the generative prompt. Example: words "Equipped for the Kingdom" + style "bold gold metallic" → 📝 To prompt → wire into Final Render.

📝 Text block — multi-line text rendered EXACTLY as typed, styled per line (deterministic, no AI, $0) → a transparent layer. Markup: # big · ## medium · > accent line · - small · blank = spacer · *word* = accent colour. Example: "# EQUIPPED / FOR THE *KINGDOM*" → a crisp gold headline, exact every time.

🔠 Text variants — wire an image in and it stamps each line of text onto it → one output per line (flat overlay = pixel-exact, or on-object). Example: a size-chart image + lines "10 in / 12 in / 14 in" → three labelled versions.

🎛 Brief — fields that steer the render

🌆 Scene — the setting/environment to generate (its own prompt section) — use when there's no background photo. Example: "a modern kitchen bench, warm morning light, shallow depth of field."

🖌️ Style — overall art direction (camera, lighting, mood, finish), a separate section from the Scene. Example: "cinematic product photography, warm tones, subtle film grain."

✍️ Prompt — the plain-English instruction / layout. Example: "product centred, man on the right, text top-left, soft shadows."

➕ Modifier — extra text appended to the Prompt (reusable / toggleable). Example: "warmer lighting."

📐 Aspect — output shape: 1:1 · 2:3 · 3:2 · 16:9 · 9:16 · 4:5 · 3:4, or a custom W×H. Non-native ratios render at the nearest size then centre-crop. Example: 16:9 for a wide banner.

🎨 Brand — apply a saved brand kit (palette + style descriptor) to the render. Example: pick your brand → its colours + look flow in.

In a generative render the Brief assembles in fixed sections: Scene → Style → Layout (Prompt) → Text.

🧱 Build — assemble & render

✂️ Cutout — remove a background, keep the real pixels → a transparent PNG (no AI, free). Example: a product photo → a clean cut-out on transparency.

🟧 Manual Compose — place your real layers exactly where you set them (no AI, never redraws). ✥ Edit layout to drag/resize, 👁 Preview to check. Example: Pete + poster + QR placed by hand.

🪄 Auto Compose — like Manual Compose but the AI arranges the layout from your Prompt brief; still real pixels (identity-safe). Example: wire 3 ingredients + "hero layout" → the AI places them.

🟩 Final Render — the render step, with an explicit mode: Auto (infer from wiring) · Generative (AI paints a NEW image from refs + prompt — not pixel-exact) · Identity-safe (finish a wired Compose — real pixels; tick ✨ AI finish to complete cut-offs + blend). Example (Generative): Scene + Text specs → a fresh poster. Example (Identity-safe): a Compose → clean layers + soft shadows.
Render history: each render is kept as a new version — a scrollable thumbnail strip (newest first) appears under the image. Click any thumbnail to bring that earlier render back into the preview; run again to add another version. History is saved with the flow, so it's still there when you reopen it.

✂️ Extract — after a render

🪓 Split Layer — pull ONE element out of a render as a transparent PNG (Person / Object / Foreground / Background, or By-prompt). Stack several to break a render into layers. Example: render → Split(Person) + Split(Background) → two PNGs.

💾 Export PSD — stack your Split Layer outputs (and/or a render) into one layered .psd for Photoshop. Example: Split(Background) + Split(Person) → a 2-layer .psd.

🎛 Reading a card (badge & status)

Engine badge (top-right of each action card) — tells you the pipeline that action will use, so you can confirm the architecture: 🟢 local · $0 = no AI, no re-render (rembg / Pillow / pytoshop — Cutout, Compose, Split Layer person/object/background, Export PSD, a plain Final Render); ⚡ subscription (or 💳 metered) = the live AI image engine (Final Render with ✨ AI finish, Split Layer By prompt, on-object Text variants). It updates live when you change a setting (e.g. ticking AI finish flips 🟢→⚡). Click an AI badge (or the top engine chip) to switch the engine: ⚡ subscription = free but slower (~minutes) · 💳 metered API = per-image $ but fast (~seconds). It's a global, admin-only setting — the switch changes it for every card and the design-bot.

While it's working — a spinning ring + “working…” shows in the card's status line, the card gets a blue outline, and the Run button greys out so you can't double-fire it. A local action finishes in seconds; an AI action (✨ AI finish) is ~4–5 min — leave it running.

If it times out or fails — the status shows ✗ with the reason and a toast pops up (e.g. “timed out — the engine is slow or out of credits”). Check the task, or press Run again.

🖱 Canvas controls

Move canvas: drag empty space (open hand). · Select: Ctrl/Cmd + drag (crosshair); Ctrl/Cmd+A = all; Esc = clear. · Zoom: mouse wheel.
Move a card: drag its body — magnetically-stacked cards move together (drag the top = whole stack; drag a lower one = it + the cards below). · Resize a card: drag its bottom-right corner. · Wire: drag an output dot → an input dot. · Disconnect: click a wire. · Enlarge: 🔍 click any generated image to open it big + Download.

✅ Typical recipe (a promo like Pete's)

+ Image ×3 — the person (Subject), the poster (Subject), the scene (Background). Pre-edit any image first if its content needs changing.
+ Compose → wire all three in. ✥ Edit layout to place them → 👁 Preview.
(optional) + Prompt/Modifier for a finishing instruction (e.g. shadows).
+ Final Render → wire the Compose in → ▶ Final Render (~1 min; cleans layers + shadows, no re-render).
+ Split Layer ×N off the render → each pulls out a transparent layer (Person / Background / Object).
+ Export PSD → wire the Split Layer cards in → ⬇ Build PSD → one layered .psd for Photoshop.

Content

Brief

Build

Extract

engine… Wire modules into a stage and Run. Select: drag empty space to box-select (Ctrl+A = all); pan: Ctrl-drag the canvas. Disconnect: click a wire, or grab an input dot to re-route. Chain a stage's output onward; Cutout keeps real pixels; Compose puts the real product on a background.

📂 Your saved flows

❓ Design Flow — what each card does

🎛 Reading a card (badge & status)

🖱 Canvas controls

✅ Typical recipe (a promo like Pete's)

✥ Layout — arrange the layers

🖌️ Mask — paint the area to change