Our AI Models

General multilingual chat (German/EU-focused)

Recommended for

General conversation, brainstorming, explanations, and multilingual chat — trained across all 24 official EU languages.

Teuken is a locally-hosted language model that can assist with a variety of tasks including:

• Answering questions
• Generating content
• Providing explanations
• Brainstorming ideas

We are using the model Teuken-7B, developed in Germany by Fraunhofer, Forschungszentrum Jülich, TU Dresden, and DFKI.

Funded by the German Federal Ministry of Economics and Climate Protection (BMWK) in the context of the OpenGPT-X project.

The model runs entirely on our local servers in Germany.

Try Teuken-7B

Coding-focused — best for code, debugging, dev tasks

Recommended for

Writing and reviewing code, debugging, explaining errors, and as a coding interface for other tools (e.g. captradeai.com).

Qwen2.5-Coder is a coding-focused language model, well suited for:

• Writing and reviewing code
• Debugging and explaining errors
• Acting as a coding interface for other tools (e.g. captradeai.com)

We are using Qwen2.5-Coder-14B-Instruct, developed by Alibaba's Qwen team, quantized to run on our small server.

The model runs entirely on our local servers.

Try Qwen2.5-Coder

Text-to-image and image editing — fast, high quality, 4-step generation

Recommended for

Generating original images from a text prompt — art, concepts, illustrations, anything you can describe — or uploading an existing image and describing how you want it changed.

FLUX.1-schnell is an open, permissively-licensed image generation model. It's a "distilled" model, meaning it produces a finished image in as few as 4 steps — most models need 20-50 — so generation stays fast even on our own modest GPU. Besides generating a fresh image from a text prompt, you can also upload your own image and describe an edit — a strength slider controls how much of the original is kept versus how far the result drifts toward the prompt.

We are using FLUX.1-schnell, developed by Black Forest Labs, Apache 2.0 licensed.

The model runs entirely on our local servers.

Try FLUX.1-schnell

Image-to-video — animate one of your generated images

Recommended for

Turning a still image into a short video, up to several seconds — pick one of your own generated images, or paste an image URL.

Wan2.2-I2V-A14B takes a source image plus a text prompt and generates a short video from it, using a two-expert (high-noise/low-noise) mixture-of-experts architecture. It's a large model, so we run it GGUF-quantized and generate in chunks internally to fit our own modest, shared GPU — videos longer than a few seconds are stitched together from several generated pieces.

What actually happens to your video, step by step:

1. Your source image and prompt go to Wan2.2's two experts (a "high-noise" pass for overall motion/composition, then a "low-noise" pass for detail), generated at 832×480 — matching the resolution the model itself was actually trained at, not a resolution we picked arbitrarily.
2. The raw output is sharpened and upscaled 2× (to 1664×960) using a dedicated image-upscaling model, to recover detail a heavily-compressed 12GB-GPU-friendly generation otherwise loses.
3. Frames are smoothed via AI frame interpolation (RIFE), doubling the frame rate (16fps to 32fps) for noticeably smoother motion, without slowing down or speeding up the clip itself.
4. Optionally, a style LoRA (see below) can be applied on top for a specific visual effect.

Every one of these steps runs automatically — there's nothing extra to configure to get the sharpened/smoothed result; picking a style below is the only optional part.

We are using Wan2.2-I2V-A14B, developed by Alibaba's Wan team, quantized to run on our small server.

The model runs entirely on our local servers.

Optional styles

• Cinematic Flare — Adds realistic cinematic lens flares when bright light hits the lens — best on scenes with strong backlighting.
• Anime Style — Converts the output to a modern HD anime art style.

Try Wan2.2-I2V-A14B