Replicate

Run and fine-tune models. All with one line of code.

Run open-source models with a cloud API, or deploy your own custom models at scale.

Colors

Primary

Black

#000000

Text, buttons

Orange-Red

#EA2804

Accent, Run

Blue

#2563EB

Links, processing

White

#FFFFFF

Page

Accent Spectrum (model tags)

Yellow

#FFE629

Tag

Pink

#FF6BFC

Tag

Magenta

#E54FE2

Tag

Salmon

#F97E82

Tag

Surfaces — Dark (neutral near-black)

Page

#0F1115

Dark background

Surface

#17191E

Cards

Elevated

#1F232B

Raised

Code / Logs

#111827

Log surface

Text & Status

Secondary

#6B7280

Muted text

Succeeded

#16A34A

Prediction done

Border (light)

#E5E7EB

Dividers

Starting

#9CA3AF

Pending status

Typography — Inter ≈ Basier Square · JetBrains Mono

Run AI with an API

40px600Inter

Thousands of models

27px600lh 1.1

Predictions & deployments

20px600lh 1.3

body

Push your model to Replicate and it becomes a scalable API endpoint — no infrastructure to manage.

15px400lh 1.6

mono

replicate.run("meta/meta-llama-3-70b-instruct")

13pxJetBrains Mono

Signature — Prediction

meta/meta-llama-3-70b-instruct Public 12.4M runs

Input · prompt

idle

Components

Buttons

Prediction Status

starting processing succeeded failed

Model Tags

text image audio

Guardrails

Keep the canvas monochrome; use #EA2804 for the Run / brand accent

Show the full prediction lifecycle and streaming logs

Set model slugs, prediction IDs, and logs in the mono stack

Show predict time and prediction ID on every completed run

Use the spectrum only for model tags/badges

DON'T

Don't make the brand blue — Replicate is monochrome + orange-red

Don't invent dark surfaces; anchor on neutral near-black #0F1115

Don't hide logs — they're how developers debug a prediction

Don't set slugs or IDs in a proportional font

Don't skip the status — a prediction always has a state