Duncan Studio — Est. 2021

We make
AI that
earns
its keep.

What we actually do

We build AI generation tools for studios, small teams, and the occasional ceramics shop in Berlin that needed to stop burning their kilns at 3am. No dashboards nobody checks. No "insights" that aren't insights.

2,400 projects shipped
180K active creators
12 foundation models
<8s avg. generation
4.9 avg. satisfaction
Established 2021 — Lagos / London / Online
What we build

Tools that do
one thing well.

We didn't add a feature roadmap because we've seen what happens when you do. Six features become sixteen. Sixteen become a product nobody finishes.

01 — Core
Image generation that doesn't
require a PhD to prompt.
Stable Diffusion XL, DALL-E 3, and Flux.1 — all in one panel. Adjust CFG scale, step count, and seed without leaving the tab. The advanced settings are hidden unless you want them.
Image · Text → Image
Real-time
Watch every diffusion step.
Steer it mid-run.
Not a progress bar. An actual view of the latent space collapsing into your image. Intervene at step 12 if it's going sideways.
Process Viz
02 — Text
Long-form writing that remembers the brief.
Text Gen
Avg. time saved
6.2h
per project, per week. Measured across 40 studios, not estimated.
03 — Music
Audio that fits the mood, not just the keyword.
Describe the feeling. Get a track. Adjust tempo, key, and density with sliders that actually correspond to something real.
Audio Gen · New
04 — Collaborate
Two people, one session. Version history you'll actually use.
Real-time co-creation without the latency of a Figma multiplayer session. Each iteration is diff-able.
Teams
05 — Marketplace
Your best prompts earn money while you sleep. The mediocre ones get filtered before they embarrass you.
Marketplace
The studio

It's supposed to feel
like a workspace,
not a SaaS dashboard.

We looked at a lot of AI generation UIs. Most of them feel like you're operating industrial equipment. Ours feels like you have a very fast intern who doesn't argue back.

Desert ruins · SD XL
Deep water · Flux.1
Forest signal · SD XL
Soft ruin · DALL-E 3
01
Write the prompt. Be specific.
Vague prompts get vague results. "A person" gets you a stock photo. "A ceramicist in her 40s, clay-dusted apron, overlit studio, overexposed window" gets you something you'd actually use. We have autocomplete that nudges you toward specificity.
02
Pick your model. They're not interchangeable.
Flux.1 for photorealism. SD XL for painterly. DALL-E 3 for when you want something that looks like a brief was actually followed. We tell you the difference upfront.
03
Watch it happen. Intervene if it's going wrong.
The generation process is not a black box here. You see every diffusion step. If it's heading somewhere terrible by step 15, you can steer it. Most tools don't let you do that.
04
Export with a license that holds up.
Every output ships with a clear commercial license. PNG, SVG, GLB, MIDI. We're not going to make you read 4,000 words of terms to find out if you can use it on a client project.
Duncan Studio — Generation
Prompt
Resolution
Size 1024 px
CFG Scale
Guidance 7.0
Ready — enter a prompt above
Encoding prompt tokens
Initialising latent space
Running diffusion
Decoding and upscaling
Done — 6.1s · SD XL
Case studies

Specific problems.
Specific results.

We don't do "transformative digital experiences." We do "the kiln no longer catches fire at 3am."

Case — 01
A 2-person ceramics studio in Berlin needed to automate their kiln firing schedules.
Tonwerkstatt Mitte · Berlin, DE · 2023

They were waking up at 3am to check temperatures. We built a text-driven schedule generator that integrated with their existing kiln controller — no new hardware, no app to learn. It runs on a single prompt per batch.

0 3am alarms since deployment
Automation · SME
Case — 02
A 60-person fashion house in Lagos needed product imagery without a photoshoot budget.
Ẹgbẹ Couture · Lagos, NG · 2024

Each season: 200+ SKUs, zero studio time. We fine-tuned a generation pipeline on their existing catalogue. Results are indistinguishable from a studio shoot. Their art director still art-directs — she just doesn't rent the space.

82% reduction in imagery costs
Fashion · Image Gen
Case — 03
A London games studio needed ambient audio for 40 levels without hiring a composer.
Greymatter Games · London, UK · 2024

They had mood boards and a vague brief. We built a music generation workflow that turned level design documents directly into audio specs. The composer they eventually did hire said the reference tracks were "unsettling in how accurate they were."

40 levels scored in 3 days
Audio · Games
Across all clients in 2024
Average time from brief to first usable output: 8 minutes.

Not first draft. First usable output. The kind you'd send to a client without apologising for it first.

2024 · All sectors
8min
Pricing

Honest prices for
honest work.

No "contact us for enterprise pricing." No hidden credit system. If you use it a lot, the middle plan makes sense. If you run a studio, the right plan is obvious.

Explorer
$0
Free. No credit card. No tricks.
  • 50 images per month
  • Standard resolution (512px)
  • Community gallery access
  • Commercial license
  • Collaboration
Studio
$89
per month · billed annually
  • Everything in Creator
  • Unlimited seats
  • Custom fine-tuned models
  • API access (100K req/mo)
  • White-label option

Stop promising.
Start shipping.

Your first 50 generations are free. No onboarding call, no "let's schedule a demo" — just the tool, working, immediately.

No newsletter. No "exclusive offers." Just the product update when something actually changes.

180,000 Creators using the platform
2.4M+ Outputs generated to date
8 min Brief to first usable output