Nano Banana models take a source picture. Describe the change, “make the background a sunset”, and choose to replace the selected image or keep it.
AI
GENERATION!
Pictures, video clips, music beds, sound effects and voice-over, from a sentence. They land on your canvas as real layers you can crop, key and export, not as files in a downloads folder.
Signed in, in the editor. AI → Generate video… opens the dialog. A description is typed — a paper boat on a rainy street, the camera following it — and the model, length, shape and resolution are shown, ready for Generate.
ZAP!
THE PROBLEM
Six tabs for one picture
You know the loop. Open the image site. Type a prompt. Wait. Download. Drag it into the design. It is the wrong shape. Back to the site. Crop, download, drag again. Now you need a five-second clip of it moving, so that is another site, another account, another credit card. Then someone asks for a music bed.
We put generation where the work already is. The prompt box opens over your file, the result lands where you are looking, and it arrives as an ordinary layer. A generated picture is an image layer. A generated clip is a video layer, or a whole scene in a film. A music bed is a clip on the timeline. Nothing is special, so nothing needs a workaround.
12
pinned image models
9
video models, 1 to 30 s
3
kinds of sound: bed, effect, voice
0
API keys to paste
PICTURES
A prompt in, an image layer out
AI › Generate image… asks for a description, an aspect ratio, an optional style and a model. Ask for one picture or up to four variants. Choose Image or SVG (editable vectors) as the output.
Click Generate and a placeholder appears on the canvas with your prompt under it. You can move it while it cooks. When the picture arrives it fills the placeholder, and from then on it is just an image layer: fill, fit, crop, mask, effects, export.
A launch post, signed in. The command palette opens Generate image; a prompt for a skincare product shot is typed, the model (Nano Banana 2) and aspect are shown, and Generate is clicked. A placeholder with the prompt appears on the canvas while the image is made, then the picture lands in it as an ordinary image layer.
EDIT, DON'T REDO
Work on the picture you already have
Generation is not only for blank canvases. Select an image and the dialog grows options for it.
REFERENCES
LOOK!Send your selection as a reference image, one per layer or all of them combined, so the model sees your layout, product or palette.
VARIANTS
x4!Right-click a generated image for Regenerate 4 variants, or regenerate it as a new layer next to the old one. Keep the winner.
CUT IT OUT
SNIP!Remove background drops the subject onto transparency, ready to sit on your own colour.
TRACE IT
VECTOR!Vectorize turns a picture into editable SVG layers. Recolour a logo sketch without redrawing it.
BLOW IT UP
ZOOM!Crisp upscale gives a sharper, bigger copy. Creative upscale adds fine detail as it grows.
MOVING PICTURES
Clips that land where you need them
AI › Generate video… takes a description, a model, a duration, an aspect ratio and a resolution, and you can ask for sound. The dialog shows an estimate in credits at the model's list price before you commit.
A clip takes from about 30 seconds to a few minutes, so it runs in the background. The dialog lists what you have started this session with its progress, and Stop cancels one. When it is done, it plays in its placeholder, which snaps to the clip's exact shape.
A launch post, signed in. Generate video is opened from the command palette, a slow push in on an app window is described, and Generate is clicked. A placeholder with the prompt takes its place on the canvas; when the job finishes the clip plays in it, at the clip’s own shape.
START FROM A FRAME
Animate the design you already made
Here is the trick that makes video generation useful for a designer. Select one or more layers and tick Start from the selection. PopCraft renders them as one picture, the same way it would export them, and the model animates from there. Describe the motion, not the scene: the scene is already yours.
To carry a clip on, select the video layer. The frame it is showing becomes the first frame of the next one. Seedance models, Veo 3.1 Fast and Grok Imagine can start from a picture; Veo 3.1 Lite works from text only.
SOUND
A music bed, cut to length
AI › Generate audio… makes three things: a music bed from a description of genre, mood, instruments and tempo; a sound effect, like a whoosh, a click or applause; and a voice-over that reads your script in one of eleven voices.
The sound lands at the playhead as a clip on the frame's timeline, or the composition's. Tick Loop and a bed plays for as long as the screen shows. Then trim it, fade it and level it in Animate like any other clip.
A launch post in Animate, signed in. AI › Generate audio… opens the dialog with Music bed picked; a mood is typed and Generate is clicked. The finished bed lands at the playhead as an audio clip on the frame’s timeline, ready to trim, fade and level.
DECKS THAT TALK
Presentation audio, all at once
A deck can get a whole soundtrack in one go. AI › Presentation audio… reads each slide's speaker notes as narration, if you want it, lays a music bed under the lot, and adds stingers between slides. The bed ducks under the voice by as much as you say.
It is mixed to 48 kHz stereo and normalised to a loudness target with a true-peak ceiling. The soundtrack is a composition, so you can trim and re-level its clips in Animate, and Present plays the same mix slide by slide. Change a slide's notes and you can regenerate just the narration that is missing or stale. Offline, it falls back to a bed and whooshes made in code, with no narration, until you can generate the real thing.
THE MENU
What is on it, really
No mystery “AI”. These are the models PopCraft offers today, by name. Which ones you can use depends on what your Hypery account serves, and the dialogs only show those.
IMAGES
- Nano Banana 2The default. Fast, follows long instructions, edits a picture, writes SVG
- Nano Banana ProStudio quality, legible text, precise edits
- Nano Banana 2 Lite · Nano BananaQuick drafts and the original
- Recraft V4.1 · Pro · Utility · Utility ProPhotoreal or flat and tidy, with vector output
- Recraft V4 · V4 Pro · V3 · V2Lighting, placed text, icons, illustration styles
Pinned at the top of the picker. Anything else your Hypery catalog serves as an image model is listed under them.
VIDEO
- Seedance 2.0 FastThe default. 4–15 s, 480p or 720p, with sound
- Seedance 2.0 · 2.0 Mini · 2.5Up to 1080p; 2.5 goes to 30 s
- Seedance 1.5 Pro · 1.0 Pro FastUp to 1080p; 1.0 Pro Fast is the cheap one, no sound
- Veo 3.1 Fast · Veo 3.1 Lite4, 6 or 8 s at 720p or 1080p. Lite is text only
- Grok Imagine Video 1.51–15 s, up to 1080p, with sound
Each model only offers the lengths, shapes and resolutions it supports.
SOUND
- Stable Audio 2.5Music beds up to 190 s, and sound effects up to 10 s
- Lyria 2 · MusicGenOther music models, where your account serves them
- AudioGenA standard tier for sound effects
- gpt-4o-mini-tts · tts-1-hd · tts-1Voice-over in eleven voices
Music and effects run as predictions; voice-over is text-to-speech.
UTILITIES
- Remove backgroundCuts the subject out onto transparency
- VectorizeTraces a picture into editable SVG layers
- Crisp upscale · Creative upscaleA sharper copy, or one with new fine detail
For the picture you already have selected.
WORDS AND LAYOUTS
Designs and copy go through the agent
Text is different. A screen is not a picture, it is frames, text, auto layout and your styles. So Generate design…, First draft (flow)…, Fill content and Suggest auto layout hand the job to the agent, with the matching skill already picked and a starter prompt you can edit.
The agent works in your real file. It places frames, writes copy that fits each text box, and sets auto layout with the same commands you would use, and each change can be undone. Its chat model comes from your Hypery catalog; the default is Claude Sonnet. Agents get the generators too: generate_video and generate_audio take the same options as the dialogs.
A launch post, signed in. The background is selected and the agent panel opens. “Make the background a deep purple” is typed and sent; the agent says what it will do, calls set_fills on the selection, and the background turns purple on the canvas. The edit is one undo step.
A launch post, signed in. The Agent rail button opens the agent panel. The headline is selected, and a request is typed in words — type the headline on letter by letter, then pop the badge in half a second later — ready to send. Manage lists the skills it can pull in, Motion design among them.
A launch post in the editor. The agent panel opens and Manage lists the built-in skills, each an instruction sheet the agent pulls in when a job matches. One is opened and Duplicate makes an editable copy of it, with its name, when to use it and its instructions.
FROM ZERO
Your first generation in six moves
- 1
Sign in
Generation runs on your Hypery account. Every request carries your own sign-in, so usage is billed to you, or to the team you have picked. PopCraft keeps no AI key of its own. - 2
Open the AI menu
It sits in the top bar. Or press ⌘K and type “Generate image”, “Generate video” or “Generate audio”. - 3
Describe it
Write what you want. For a video, say what happens and how the camera moves. For a picture you are editing, describe the change. - 4
Pick the model and the shape
Aspect, size, variants, length, resolution, sound. The dialog only offers what the chosen model can do, and shows an estimate in credits where there is one. - 5
Click Generate, keep working
A placeholder holds its spot on the canvas with your prompt under it. Move it, carry on designing, and the result lands in it when it is ready. - 6
Treat it like anything else
An image is an image layer. A clip is a video layer or a scene. A bed is an audio clip. Crop it, mask it, key it, export it.
THE OLD WAY
Generate, download, drag, repeat
| THE OLD WAY | IN POPCRAFT |
|---|---|
| A different site and account for images, video and music | One AI menu, one sign-in, one credit balance |
| Download, drag in, find it is the wrong shape | Pick the aspect first; it lands in place, at its shape |
| Describe your layout in words and hope | Start from the selection: the model sees your design |
| Flat PNGs you cannot recolour | SVG output, or Vectorize, as editable layers |
| Scrub a music track to fit in another app | A bed on the timeline, looped, trimmed and faded |
FINE PRINT
What to know before you press Generate
IT COSTS CREDITS
Generation uses your Hypery credits. Estimates are based on list prices; the real charge comes from your Hypery account. A 5-second 720p clip is about 10 credits on Seedance 1.0 Pro Fast and about 60 on Seedance 2.0 Fast.
IT NEEDS A CONNECTION
Signed out or offline, the AI menu says so instead of failing halfway. Everything you have already generated stays in the file and keeps working offline.
STOPPING MAY STILL COST
Stop cancels the wait on a video, and so does deleting its placeholder. A model that has already started may still charge for it.
MODELS COME AND GO
The list follows what Hypery serves. If a model is not offered on your account, it is not offered in the dialog, and a sound model nobody serves says so.
TRY IT NOW
One sentence, one layer
Open any template, sign in, and press ⌘K. Type Generate image, describe the picture you were about to go and find, and drop it straight into the layout. The video docs cover every model and option.
ASKED & ANSWERED
Questions, answered
Which AI models does PopCraft use?
For images, Google’s Nano Banana family and Recraft V2 to V4.1, plus any other image model your Hypery catalog serves. For video, Seedance (2.5, 2.0, 2.0 Fast, 2.0 Mini, 1.5 Pro, 1.0 Pro Fast), Veo 3.1 Fast and Lite, and Grok Imagine Video 1.5. For sound, Stable Audio 2.5, with Lyria 2, MusicGen and AudioGen where your account serves them, and OpenAI text-to-speech for voice-over. The agent’s chat model comes from your Hypery catalog too.
Do I need my own API keys?
No. You sign in and generation uses your credits. Requests go out with your own sign-in, so there is no shared key, and no key of yours to paste anywhere.
Can I generate vector art?
Yes. Pick SVG as the output in Generate image: Recraft models draw it, and Nano Banana writes SVG markup. It lands as editable vector layers, not a flat picture. Vectorize can also trace an existing picture into SVG.
Where do generated files live?
In the file’s local media, like anything you drag in: saved on your device and synced for cloud files. They are never embedded in the document itself, so your file stays small.
Can the AI design a whole screen?
Generate design and First draft (flow) hand the job to the agent, with the matching skill picked. It builds in your real file, with your styles and components, and you can watch and undo each step. It is not a one-shot picture of a screen.
What happens if a generation fails?
The placeholder turns into a red card that says why, such as a prompt the model refused or running out of credits. Stopping a video, or deleting its placeholder, cancels the wait, though a model that has already started may still charge for it.
THE MANUAL, FOR THE DETAILS
ZAP! YOUR TURN.
Images, video, sound and layouts from a prompt, on your canvas. Free to start, in your browser or on your Mac.