A
adstudioh
Talk through the brief. Rough thoughts in — clarifying questions and a tightened brief out.
Working brief
Updated as you talk — edit freely, then approve
Keep talking until the brief feels right, then approve before Make.
References & uploads
Optional · drop 1–3 reference images, or tap to upload
More direction
Sessions stay private. Collections and Library are how work becomes shared.
Analyzed once, saved on this device, fed into every generation.
Explore
Brief in → a living stream of visual directions. Scroll for more. React as you go. Lock format only when you pick a winner.
Approved brief
No approved brief yet. Finish Talk, approve the brief, then come back here.
Fast cheap Explore model · mixed shapes · no platform sizes yet. Scroll near the bottom for more; stop scrolling to pause.
Stream → Finalise
- Scroll the feed — mixed shapes and moods appear as you go
- Shortlist / More like this to bias what comes next
- Open a card for a closer look, then Finalise into LinkedIn / Meta / Story
- Convert to edit live text on the template
Direction
Library
Saved ads and references in one visual gallery. Like winners, open them in the Editor, save as anchors, or file into collections.
Add a reference
Drop an image here, or use the button below
![]()
Collections
Group ads and references by campaign or theme. An item can belong to more than one collection. Mark a collection shareable for a feedback link.
No collections yet. Create one above, then add items from the gallery.
Add to collection
Choose a collection for this item.
Settings
Configure OpenAI image generation and edit House Rules — the single source of truth for brand identity and voice. Every Generate and AI assist reads the latest rules.
Appearance
Auto follows your system setting. The tool chrome stays neutral; Foam lime and brand colours appear on ad creatives only.
Admin · Login splash
Full-bleed image behind Google sign-in. Visible to everyone on the login gate. This control is admin-only — invited teammates will not see it.
No file chosen
OpenAI · image generation
Talk, Make directions, and AI assist use gpt-5 via the Cloudflare worker (fallbacks: gpt-4.1 → gpt-4o). Make is Explore-first (fast cheap image model, mixed shapes, no platform size yet). Finalise locks LinkedIn/Meta/Story size with gpt-image-2. Images upload to R2. The API key must be set as worker secret OPENAI_API_KEY — nothing secret is stored in Settings.
House Rules · Voice & Campaign
Single source of truth for voice. Generate and AI assist read these fields — nothing important should live only in a hard-coded prompt.
Brand identity · Visual system
Palette, lime, type, layouts, and proof primitives. Edit here — the Editor panel and generation prompts pull from these fields.
Edit a shortlisted winner
In Make, Explore directions, shortlist a winner, Finalise into a real format, then Convert. You get the template, a text-free background, and live editable layers — refine the words.