The ✦ verbs never hide in a menu: the chips above the AI box follow the selection, and the ✦ pill under the selection shows up to five of the same. This page is the complete list. Read The AI assistant first for how chips, the box and History fit together.
01Verbs by selection
| Selection | Chips |
|---|---|
| Nothing | Make — a new image from a prompt (and Ref, if set) · Make video — text to video · Make sound — Music, SFX or Speech, lands as a timeline track. With Design set to Google Stitch, Design screen comes first — see Importing HTML & Stitch screens |
| An artboard | Paint (render the blockout in place) · Take (a new angle beside it, camera-aware — the chip counts up: Take 2, *Take 3*…) · Make |
| An artboard with a 3D scene inside (a set) | Take, Paint, and World — generate an HDRI and light the set with it |
| An image | Change (say what changes — "make it red", "replace the label with NOVA" — everything else stays; on a take, Apply to all N takes changes every angle at once) · Vary (four takes on the same result) · Animate (turn it into a clip) · Edit ▾ (below) · Use as Ref · Build 3D (a 3D model in a new set beside it) |
| A video clip | Extend (a new clip continuing from its last frame) · Restyle (re-grade it in a new look) · Frame → (grab its current frame as a new image beside it) |
| A 3D mesh in a set | Retake — a new take of the mesh from its set, from the current camera angle |
| Two images, or two takes | Between (a clip from one to the other, camera-aware) · Blend (the midpoint of both) |
| A text layer | Illustrate (the text becomes the prompt) |
Use as Ref and Frame → run the moment you click them. Every other chip arms: type what you want, then Enter or the Send button, which now carries the verb's name. Esc disarms. A chip greyed out has no model available yet; a dot on a chip marks a verb whose models are all still beta.
Every verb that supports it sends ✦ Ref automatically and leans on ✦ Look — see Reference, Look and Recipe.
02Edit ▾
Edit ▾ on an image opens a row of actions. Each one runs when you click it — it does not arm:
- Remove — paint over the object you want gone. A red brush ring under your pointer shows exactly what will be covered. Paint the whole object plus a little margin, then Apply — nothing is sent until you do. No description needed.
- Erase & fill — the same brush, but you type what should appear in the painted area: "a wooden bench", "more of the same sky". Paint, describe, Apply.
- Upscale — sharper and larger, no prompt.
- Cut out — removes the background, leaving the subject on transparency.
- Parallax layers — splits a photograph into depth-ordered, transparent layers with the background filled in behind what was in front — ready for a scroll parallax.
- Split into layers — separates an ad- or poster-style image into its editable pieces: background, product or logo cutouts, live text, simple shapes. Reads a flat composition.
- Use style as Look — pulls the image's own palette and mood straight into ✦ Look, no model call, no brush.
03Options (▾ next to Send)
With a verb armed, ▾ opens that verb's options:
| Option | For | Choices |
|---|---|---|
| Quality | Every verb except Design screen | Fast or Best (slower) |
| Count | Make, Illustrate | ×1 or ×4 |
| Aspect | Make, Illustrate, when Ref is empty | 1:1 through 9:16 |
| Layout | Paint, Take, Retake | Loose, Follow or Tight — how closely the result keeps your blockout |
| Model | Every verb with a choice | Auto, or a named model with its price |
| Device | Design screen | Desktop, Mobile or Tablet |
The last line says how many Ref images and whether Look is on. Advanced form… opens the full generation form — any model, negative prompt, seed, every parameter — in place of the box; Back to chat returns.
The foot of the box shows the estimate before you press anything — ~$0.04 · Flux dev — and ✨ rewrites a rough prompt for the verb and model in use. Pick Claude or Gemini for ✨ under Settings › AI & Generation › Prompt enhancer provider.
Every result lands as a ghost over its target: Enter keeps it, Esc discards it, ← → walk the variants, Shift+Enter asks for four more. Until you keep, nothing has touched the scene.
04Which model
Each verb picks from a short list of real models. Auto picks a fast one; Best (slower) picks the better tier; Model names one yourself. Costs and timing show next to each name in the picker and in a result's Why — they change as models do, so numbers aren't repeated here.
| Verb | Fast | Quality / alternatives |
|---|---|---|
| Make (no Ref) | Flux Schnell, Qwen Image | Flux Dev, Flux Pro 1.1, Nano Banana, Seedream 4, Imagen 4, Ideogram 3 (typography/posters), Recraft 3 (vector/illustration), Nano Banana 2 and Nano Banana Pro (through your Gemini key) |
| Make (with Ref), Change, Blend | Nano Banana edit, Qwen edit | Flux Kontext, Seedream 4 edit, Kontext Max, Nano Banana 2 edit and Nano Banana Pro edit (through your Gemini key) |
| Paint / Take (with Ref) | Nano Banana edit | — |
| Paint / Take (no Ref) | Flux dev i2i | — |
| Vary, Illustrate | Same pool as Make, with or without Ref | |
| Edit › Upscale | AuraSR | Clarity upscale, SeedVR upscale |
| Edit › Cut out | BiRefNet | Bria BG remove |
| Edit › Remove, Erase & fill | Flux Fill | — |
| Edit › Split into layers | Bria Ad Delayer | — |
| Edit › Parallax layers | Qwen Image Layered | — |
| Animate, Between, Make video | Kling 2.5 turbo, Wan 2.2, Hailuo 02 | Kling 3 pro, Luma Ray 2 (first+last frame), Veo 3 (sound-synced) |
| Extend | Veo 3.1 extend | — |
| Restyle (video) | Kling O1 restyle | — |
| Build 3D | Hunyuan3D 2 | Trellis, Tripo 2.5, Meshy v7 |
| World | Hunyuan World | — |
| Make sound › Music | Stable Audio 2.5 | — |
| Make sound › SFX | ElevenLabs SFX | — |
| Make sound › Speech | ElevenLabs TTS | — |
| Design screen | Google Stitch — your Stitch quota, no per-shot price | — |
A model marked beta in the picker is wired in but hasn't been run for real by this studio yet.
05From elsewhere
The same verbs open from other places, always arming the chip in the panel:
- Right-click an image → ✦ Change…; an artboard → ✦ Render… / ✦ Take… (Shift+R); two images → ✦ Video between these.
- Properties › Generated on a generated image → Regenerate, Vary ×4, Change…, Advanced… — see History and Storyboard.