2pm.spaceMagic Studio:
on-brand AI images
Put a product photo, a prompt and any of 25 image models on a canvas of steps you can run again. Your packshots come along whenever the prompt names the product, and a rough note is rewritten into a proper prompt from your brand.
From a product photo to a campaign visual
Three steps, and every one of them can be run again on its own.
Drop in what you have
A product photo from your computer, from Drive or from the brand's own pictures, and a prompt — as rough as you like. Each can be a node of its own, or a reference on the image step itself.
Wire the steps
Wire the prompt into Generate or a photo into Edit Image, pick a model for each step, and set the ratio, resolution, quality and how many to make.
Rerun until it is right
Re-generate lands beside the last result instead of over it, so every attempt stays on the canvas. Cut the background out, edit a result further, or download it.
Drop in what you have
A product photo from your computer, from Drive or from the brand's own pictures, and a prompt — as rough as you like. Each can be a node of its own, or a reference on the image step itself.
Wire the steps
Wire the prompt into Generate or a photo into Edit Image, pick a model for each step, and set the ratio, resolution, quality and how many to make.
Rerun until it is right
Re-generate lands beside the last result instead of over it, so every attempt stays on the canvas. Cut the background out, edit a result further, or download it.
Built for images you will need again
Not a prompt box — a canvas that remembers how every picture was made.
Twenty-five models on one canvas
GPT Image, Seedream, FLUX.2, Nano Banana, Qwen Image, Ideogram, Recraft and more — each step picks its own, so comparing two models is two nodes side by side rather than two browser tabs and a spreadsheet.
Your products, on-model
Every generator step carries a Brand. Name one of its products in the prompt and the packshots you uploaded come along as references, so the bottle on the poster is your bottle.
A prompt writer that knows the brand
Write a rough note and the art-director agent turns it into a proper English prompt from your brand profile and catalogue, and marks which products it is about. It lands in the prompt box before anything is generated, ready to edit.
Edit with up to 16 references
Change what is in an image with a sentence and a handful of references — the product, the setting, the model, the palette — and keep the source's size when it matters.
Every attempt kept, cut-outs included
Re-generate adds a new node beside the last one with the same settings, so earlier results stay to compare. Remove the background of a result — the cut-out is added next to the original, ready for packshots and marketplace listings.
Made together, traceable later
Your team edits the same canvas in real time, and every image keeps a record of the steps and sources it came from — so the one that worked can be made again.
Models, steps and output
The whole list, before you decide whether it covers what you make.
Models
- OpenAI — GPT Image 2 and 2.5
- ByteDance — Seedream 5 Pro and 4
- Black Forest Labs — FLUX.2 and Flux
- Google — Nano Banana, 2 and Pro
- Alibaba, Ideogram, Recraft, HiDream, Krea, Bria, ImagineArt
Steps
- Generate — one to four images per run
- Edit Image — a prompt plus up to 16 references
- Re-generate — a new node beside the last, same settings
- Remove Background on any result
- Image input from upload or Drive
Output
- 1:1, 3:2, 2:3, 16:9, 9:16, 4:3, 3:4 and 21:9
- Or a custom size from 256 to 4096px
- 1K or 2K resolution
- Quality from Draft to Max
- A fixed seed when you need a repeatable result
Brand
- A Brand per generator step
- Product packshots attached when the prompt names the product
- Prompt rewrite from the brand profile and catalogue
- Rewrites land in the prompt box, ready to edit before you generate
On every result
- Remove the background
- Edit it further
- Download it
- Kept in the image library, ready to pick for a post
Team
- Real-time co-editing on one canvas
- Source and step history on every image
- Canvases kept in Drive with the rest of your work
- Usage counted per workspace — no provider keys
What a creative team stops doing
The difference between a prompt box and a studio.
A subscription per image model, and a folder of downloads to compare them in
Twenty-five models on one canvas, one workspace bill
A lucky prompt nobody can reproduce once the chat history is gone
A canvas of steps you can rerun, with every image's sources recorded
A product that comes out a slightly different colour and shape every time
Your packshots attached as references whenever the product is named
Prompts written by whoever is best at prompting, and nobody else
A rough note rewritten by an agent that knows the brand
Images passed around on chat, then re-uploaded into the scheduler
Every result in the image library, ready to pick for the post it was made for
Questions about Magic Studio
What teams ask before they move their product shots over.
Twenty-five, from eleven makers: GPT Image 2 and 2.5 from OpenAI, Seedream from ByteDance, FLUX.2 and Flux from Black Forest Labs, Nano Banana from Google, Qwen Image from Alibaba, Ideogram, Recraft, HiDream, Krea, Bria FIBO — trained on licensed data — and ImagineArt. Each generator step on the canvas picks its own model, so one canvas can compare several.
No. Every model runs through the platform, so there are no provider keys to collect and no separate bills. Generation is counted as usage on your workspace, like the rest of the AI you use here.
Each generator step carries a Brand. When the prompt names one of that brand's products, its reference pictures — the packshots you uploaded — are attached to the request automatically, so the model works from the real item instead of guessing at it. Edit Image takes up to 16 reference images of its own for everything else.
Less than you would think. Write a rough note and the art-director agent rewrites it into a full prompt from the brand profile and the product catalogue — in English, which image models read best — and marks which products it is about. The rewrite goes straight into the prompt box before anything is generated, so you can still edit it or write over it.
Video lives in Storyboard, where a shot has the camera, the action and the characters it needs to be rendered well. Panels drawn there use the same image models as Magic Studio.
Yes — canvases are edited together in real time, and every image keeps a record of the steps and sources it came from. Finished images can be downloaded, or picked from the image library for a post on the content calendar.
Make this week's visuals on one canvas
Upload a product photo, write a rough idea, and compare three models side by side. Free to start, no credit card.