Marketing Studio

The Marketing Studio is where you make the visuals. Describe what you want, pick a model, and get images back - with the exact credit cost shown on the button before you click it.

It sits at /marketing-studio in the sidebar. It has its own workspace: a composer, a project canvas, an Inspiration library, and a set of AI influencers you can use as presenters.

The four surfaces

Opening the Studio always drops you into a blank session - a clean composer with no project attached yet.

Generate images

This is the path that is fully wired end-to-end today.

1. Open /marketing-studio. You land on Creation, in a blank session.
2. Switch the mode toggle to Image (the Generate button turns black/white; video mode is red).
3. Write your prompt. The box grows as you type, up to 8 000 characters.
4. Open the preferences popover and choose:
   • Model - five image models are available, from 30 credits an image to 370 credits an image. The top model renders at a larger native size (2K); the rest render at 1K. Price tracks the model's real cost, not its resolution - the cheaper ones are for volume and iteration.
   • Ratio - 9:16, 3:4, 4:3, 16:9 or 1:1.
   • Images - a stepper from 1 to 10.
5. Optionally attach assets (see below).
6. Read the number on the Generate button. That is what you will be charged.
7. Click Generate. The button returns immediately - the job now runs on our servers.

Attach what the image is about

Three asset slots sit under the prompt:

• Product (one) - pull a product straight from your Neural Core, or upload one from your device.
• Influencer / presenter (one) - one of the eight presets, one of your saved avatars, a photo you upload, or one you generate from a description.
• Reference images (several) - from your image library or your device. Drag them to reorder: the order maps to @Image1, @Image2… inside the prompt that gets compiled, so you can refer to them positionally.

Type @ anywhere in the prompt to mention a product or a library image and pull it into the generation without leaving the box. Mentions become chips you can hover to preview and click to re-route.

Uploads accept PNG, JPEG, WebP, GIF and AVIF, up to 12 MB each, with a real progress bar.

Your draft is saved per project

Everything in the composer - the prompt, the product, the presenter, every reference image, the model and all the settings - is autosaved per project. Leave, go answer an email, come back a day later: it's exactly as you left it. A draft you started in a blank session carries into the project that your first generation creates.

Projects behave like conversations

There is no "New project" button, and that's deliberate.

• The first generation of a blank session creates the project. It's named instantly from your prompt, then a model quietly upgrades the title in the background.
• If you rename it while that's happening, your name wins. The background re-title never clobbers a manual rename.
• A project appears in the sidebar the moment its first generation starts, with a spinner.
• It persists once media lands.
• It quietly disappears if that first generation failed and left it empty - no debris, no empty folders to clean up.

Rename, favourite or delete a project from the sidebar's "…" menu or from the Projects grid. Favourites sort to the top.

Close the tab. The job still finishes.

When you click Generate, the Studio reserves your credits, creates the row and returns a generation id in milliseconds. The actual work continues in the background, on the server.

Close the tab. Log out. Lose your Wi-Fi on a train. Never come back to the page. The media still lands in your project, and your credits still settle correctly - because nothing about finishing the job depends on your browser being open.

Two independent processes settle every generation: the background task and a sweeper that runs every two minutes. Exactly one of them can win - the settlement is a compare-and-swap on the row, so you can never be double-charged, no matter which one gets there first.

How the money works

The Studio spends the same credits as everything else in Unyo.

The price on the button is the price you pay

The credit label in the composer and the server's actual charge run the same formula from the same database rows. Nothing is hardcoded on either side.

• Images: rate × number of images.
• Video: rate × duration × resolution multiplier.

Because the rates live in configuration rather than in code, pricing can be adjusted without a deploy - and the composer picks up the new number immediately.

Partial batches are billed pro-rata

Ask for 10 images and get 7? You are charged for 7. When a batch finishes, the commit is recalculated as charged × delivered ÷ asked, not the full reservation. You pay for pixels you received.

Failed work is refunded automatically

Every failure path refunds. And if a generation dies in a way that leaves a reservation stranded - the server isolate was killed, you never came back, the provider hung - a safety net scans for reservations still open after 10 minutes and settles them against what actually happened: committed if the media landed, refunded if it didn't.

Neither credits nor status can be forged from the browser

The generations table gives you read and delete on your own rows and nothing else - there is no insert or update policy for a signed-in user at all. Every write to status, outputs and credits goes through the server. Your balance is not a client-side number.

Video, honestly

There is a video mode in the composer - model choices, durations, resolutions, aspect ratios and creative formats, wired to the same durable-generation and refund machinery as images. We're not documenting it as a live capability, and we're not going to quote its price grid: no customer video has been generated end-to-end in production yet. Judge the Studio on images, which are live today; when video ships for real, this page will document it properly.

What the composer already shows you, for the curious:

• Three models, in a speed/cost ladder.
• Duration: 4, 5, 6, 8, 10, 12 or 15 seconds - a fixed set, chosen with a scrubber. Default 5.
• Resolution: 480p, 720p or 1080p, depending on the model. Default 720p. 1080p is only on the top model.
• Aspect ratio: 16:9, 9:16, 4:3, 3:4, 1:1 or 21:9. Default 9:16.
• Creative format: None, UGC, Tutorial, Unboxing, Hyper Motion, Product Review or TV Spot.

A format is not a template video. It selects a prompt recipe - camera language, lighting, pacing, structure beats and do/don't rules - that a model uses to rewrite your brief into one cinematic prompt. If that rewrite fails for any reason, your raw prompt is used as-is rather than the request being dropped.

Attached assets change the generation, and the price:

• One image → it is used as the first frame.
• Two or more → they are used as identity-preserving references.
• Any product, presenter or reference image at all flips the charge from the text-to-video tier to the reference-to-video tier - visibly, on the button, before you commit.

Inspiration and Recreate

Inspiration → Video is a curated shelf of video ideas. Cards play looping previews and unmute when you hover.

Inspiration → Image is the community prompt gallery - the same gallery available standalone at /gallery, embedded here. It spans ten categories (portrait, product, lifestyle, abstract, art, fashion, food, nature, architecture, plus your saved prompts) and loads 24 at a time.

The recommended feed is genuinely scored, not shuffled: a sector match on your business is worth +30, affinity with the categories you already use +20, popularity up to +15, freshness +10 if it's under a week old, with a −5 penalty for seeing the same author twice in a row and a randomised factor so the feed doesn't calcify. It's scored once and paginated in memory, so scrolling doesn't reshuffle under your thumb.

You can like, save, copy and publish prompts. Publishing runs your brand name and your product names out of the prompt text first - you're sharing the technique, not your roadmap.

Recreate

Recreate on any card is a real deep-link, not a copy-paste.

• On a video inspiration, it rehydrates the prompt, model, ratio, resolution, duration, format, product, presenter and every reference image into the composer.
• On a gallery prompt, it opens the composer in image mode with the prompt and its reference images pre-filled.

Either way you land in Creation with a loaded composer, one edit away from your own version.

AI influencers

Eight ready-made presenters cover the usual niches - beauty and lifestyle, fashion, tech and gaming, wellness and fitness, K-beauty, food and travel, sports and outdoor, music and entertainment. Filter by All / Favourites / Female / Male, and favourite the ones you keep reaching for.

Need someone who doesn't exist yet? Open the Avatar Playground and build them from chips across nine attribute groups:

The prompt compiles live as you click, so you can watch the description assemble itself. Generate the avatar, or just copy the prompt and take it elsewhere. You can also skip all of it and upload a photo of a real person as the presenter - yours, or one you have the rights to.

Troubleshooting

A generation failed. Your credits were refunded. If not immediately, the sweeper settles it within minutes. Try a different model - models fail for different reasons - and simplify the prompt.

A prompt was rejected on safety grounds. Model providers block some content. Rephrase, or drop the reference image that's triggering it.

A video seems stuck. Long video jobs are handed off from the initial request to a background sweeper that runs every two minutes, so status can lag behind reality by up to about two minutes. Give it that long before assuming it failed. If it truly stalls, it is timed out and refunded automatically.

I got fewer images than I asked for. That's the pro-rata path working. You were charged only for the ones that landed. Re-run for the rest.

My project vanished. If its only generation failed, the project is removed rather than left empty. Nothing else deletes a project on its own.

The composer forgot my setup. Drafts save per project. A draft written in a blank session carries into the project created by your first generation - but a draft is not shared between projects.

FAQ