Visuals

The Designer is a built-in worker that turns written briefs (“a minimalist logo for a coffee bar”) into actual images and videos. It can use several generation services — this tab is where you decide which ones are switched on and the order Designer tries them in.

Keys live on Providers

This tab is only about priority and preferences. The actual API keys live on the Providers tab: Gemini “Nano Banana” uses your Gemini key, OpenAI image generation uses your OpenAI key, and Replicate reads a REPLICATE_API_TOKEN from your environment. One source of truth for keys, no duplication.

How priority works

Each section (images and videos) is an ordered list. Designer tries the top-most enabled provider first; if that one isn’t available, it falls to the next. A coloured stripe on the left edge of each card shows whether the provider is fully wired in (green) or still a planned stub (grey).

🖼 Image generation

The image providers, in default order:

1. Gemini — Nano Banana — Google’s gemini-2.5-flash-image. Strong at stylised and illustrative work. Uses your Gemini key.
2. OpenAI — gpt-image-1 / DALL·E 3 — strong at photoreal images and layouts that include text. Uses your OpenAI key.
3. Replicate — Flux / SDXL / any model — runs essentially any open image model by its id. Uses a REPLICATE_API_TOKEN.

🎬 Video generation

The video providers, in default order:

1. Replicate — LTX / Hunyuan / any video model — the working video provider today. Uses a REPLICATE_API_TOKEN.
2. Seedance (ByteDance) — planned. Shown as a stub until the integration lands; selecting it returns a clear “not yet” error.
3. Dreamina (ByteDance) — planned. Also a stub for now.

The controls on each provider card

Every card — image or video — has the same controls.

↑ / ↓ (priority arrows)

• What they do: move a provider up or down the list. Higher = tried sooner.
• Default: the order listed above.
• When to change it: put the provider you trust most (or that costs least for your kind of work) at the top. For example, move Gemini above OpenAI if you mostly make illustrations, or the reverse for photoreal product shots.

Enabled (checkbox)

• What it does: lets Designer use this provider at all. Disabled providers are skipped entirely.
• Default: on for the wired providers; off for the planned stubs.
• When to change it: turn one off to stop Designer ever reaching for it — for instance, disabling Replicate if you’d rather not spend its credits this month.

Daily cap (number)

• What it does: an intended ceiling on how many generations per day this provider may do.
• Default: 0, which means unlimited.
• When to change it: set a number to remind yourself of a self-imposed budget per provider.

The daily cap is informational for now

At present the daily cap is a note to yourself, not an enforced limit — the counter that stops Designer at the cap is still being wired up. Treat it as a budgeting reminder, not a hard stop.

Note (text)

• What it does: a free-text scratchpad on the card (e.g. “used ~$5 in Replicate credits this week”).
• Default: empty.
• When to change it: jot anything you’d want future-you to remember about that provider.

Where Designer’s output goes

Every generated image or video is saved automatically to your Library under library/visuals/, with a small sidecar file recording the prompt, which provider made it, and when. Open the 📚 Library to browse the grid and click any tile to enlarge or play it.

Override per request

These settings are Designer’s defaults. You can always override them for a single request by naming the provider when you ask — e.g. “Designer, make this with OpenAI.”

In practice

Order the list by what you actually make. The priority arrows decide which generator Designer reaches for first, and the providers genuinely differ: Gemini’s Nano Banana is the stronger choice for stylised and illustrative work, OpenAI’s image model for photoreal shots and anything where text has to render legibly inside the image. Put the one matching your usual output on top and you stop having to name a provider in every request.

What this makes possible day to day:

• Social and blog imagery without a design round-trip. Brief Designer the way you’d brief a freelancer — subject, style, aspect — and the result lands in the Library ready to use.
• Consistent variations of one concept. Because every output saves a sidecar with its exact prompt, you can reopen what worked and re-run it with one detail changed, instead of trying to remember how you phrased it.

Enabled is your spending control here, not the daily cap. The cap is currently a note to yourself — it isn’t enforced — so if you want to stop spending Replicate credits this month, untick Enabled on that card. That actually prevents the calls.

Everything lands in the Library automatically, under visuals/, each file with a sidecar recording its prompt, provider and timestamp. That record is what makes a good result reproducible later, so resist deleting the sidecars when tidying up.

Stubs are honest placeholders. The grey-striped cards (Seedance, Dreamina) aren’t wired yet and return a clear “not yet” error if selected — they’re listed so you can see what’s planned, not so you can use them today.

Terms in this page

• Designer — the built-in worker that turns written briefs into images and videos.
• Provider (visual) — a service that generates images or video. Different from your LLM providers, though some share the same key (Gemini, OpenAI).
• Priority order — the sequence Designer tries providers in; the top enabled one goes first.
• Stub — a provider that’s listed but not yet functional; it’s a placeholder for a planned integration and returns a clear error if used.
• Nano Banana — the nickname for Google’s gemini-2.5-flash-image image model.
• gpt-image-1 / DALL·E 3 — OpenAI’s image-generation models.
• Replicate — a service that can run almost any open image or video model by its id, billed by usage via a REPLICATE_API_TOKEN.
• REPLICATE_API_TOKEN — the environment variable (a setting in your computer’s shell) that holds your Replicate key.
• Daily cap — an intended per-day limit on generations (currently a reminder, not enforced).
• Library — the in-app store of files your workers produce; generated visuals land in its visuals/ area.
• Sidecar file — a small companion file saved next to each output that records its prompt, provider, and timestamp.
• Prompt — the written description you give a generator to produce an image or video from.