GGenAI Studio
Sign in
Serverless GPU pipeline

Generative media,
billed by the second.

Edit images and clone voices on dedicated GPUs that spin up on demand and shut down when idle. Every run is tracked to the fraction of a cent, so you always know what a generation cost.

Open the studioHow it works
2
live modules
1
base URL
~$0.02
per generation
Abstract neural network visualisation
Generating80%

What it runs

Abstract coloured light patterns
Image

Kontext

Compose up to four reference images and reshape them with a prompt. Runs FLUX.1-Kontext on an A100.

Prismatic light refracting through glass
Image

Flux.2

Generate from a prompt alone, or steer it with up to six reference images kept separate at full resolution. Runs FLUX.2-dev on an H100.

Geometric light structures
Image

Qwen Text-to-Image

Generate at 2K from a prompt, or drive the geometry with real pose, depth and edge control. Apache-2.0 licensed and open weights.

Layered translucent surfaces
Image

Qwen Edit

Edit with up to three references the model actually reads, under stackable skin realism and multiple-angle adapters. Apache-2.0 licensed and open weights.

Audio mixing console faders
Speech

Chatterbox

Clone a voice from a short reference clip and make it read any script. Runs on a cost-efficient T4.

Audio mixing console faders
Speech

IndexTTS-2

Clone a voice from a short clip, then steer the emotion from a second clip, an emotion mixer or a text description. Non-commercial licence.

Audio mixing console faders
Speech

Seed-VC

Turn an existing recording into your reference voice while keeping its timing and delivery. Speech or singing.

Audio mixing console faders
Speech

Step-Audio-EditX

Edit an existing spoken clip: change its emotion or style, speed it up, add laughs and breaths, denoise or trim it. Apache-2.0 licensed.

Layered translucent surfaces
Video

Wan 2.2 S2V

Turn a character image and a speech recording into a talking video, with the length following the audio. Apache-2.0 licensed and open weights.

Layered translucent surfaces
Video

Wan 2.2 Animate

Make a character copy the motion of a reference video — dances, walking and gestures — and keep the original audio. Apache-2.0 licensed and open weights.

Layered translucent surfaces
Video

LongCat Avatar

Lip-synced talking avatar video from an image, speech and a scene prompt, extended in segments for longer speech. MIT licensed.

Layered translucent surfaces
Video

LTX-2.5

Generate a short video with synchronized sound from a prompt, or animate a first-frame image. Distilled 22B model; free under the LTX community licence for small creators.

Layered translucent surfaces
Video

MiniMax H3

Cinematic clips up to 15 seconds with synchronized stereo sound, from a prompt or a first frame. A 33B open-weights model run across four GPUs.

Abstract coloured light patterns
Enhance

SeedVR2

Restore and upscale a finished image or short video, keeping the original audio. Choose a 3B or 7B model. Apache-2.0 licensed and open weights.

Abstract coloured light patterns
Enhance

SUPIR

Rebuild fine detail in a low-quality still image with generative restoration, steered by a short description. Non-commercial licence.

How it works

01

Request hits the gateway

A single base URL fronts every module. The CPU gateway validates, writes a job row, and returns an id in milliseconds.

02

A dedicated GPU picks it up

Each module spawns its own worker with its own GPU class, so speech synthesis never pays for an A100 it does not need.

03

Progress streams back live

Workers write progress straight to the database and the browser subscribes to that row. No polling, no waiting on an open connection.

Network of illuminated connection points

Know what every run costs

GPU, CPU and memory time are recorded per second and attributed to the project that triggered them — including the idle minutes and cold starts most platforms quietly bill you for.

Open the studio