Search the full model catalog.

Tencent’s Hy-MT2-7B is an open-weights multilingual translation model for 30+ languages. It follows glossary, style, and format instructions for faithful output.

EMA Lightning turns Turkish text into speech and can stream mono WAV audio up to 48 kHz. Adjust speaking speed and set a seed for repeatable takes.

Qwen-Image-2.1 is Qwen’s open-weight 7B model for text-to-image and image editing. It supports 2K PNG output, including native RGBA transparency.

FastH3 V2 is an eight-forward distilled MiniMax H3 model that generates synchronized video and stereo audio from a text prompt. This H200-backed 8-step checkpoint supports native 5.167-14.375 second T2AV clips at 24 FPS; FL2VA and Ref2VA were not distilled.

Builds a short video from one portrait. The person faces their own double at an edge, such as a canyon, quarry or rooftop, gets pushed, and the camera follows the fall.

Transform a single car photo into a cinematic vehicle adventure. Turn your car into a powerful robot, a stylized cartoon character or place it inside an iconic open-world street scene — while preserving its recognizable design and details.

Turns one portrait into a rescue video: the person is in danger, for example on a sea rock, in ice, sand, floodwater or a burning room, and a second version of them pulls them out. All eight scene previews end in a hug.

Turns one photo of a person into a short video set in a preset 1980s scene, such as Arcade Night, Diner Night, 80s Proposal or 80s Family, with period outfits and hair. There are 16 scenes to pick from.

Turns one portrait into a multi-shot video of the person cutting their own hair, from the first snip to the reveal, in one of 12 cuts such as a blunt bob, pixie or mullet.

Animates one portrait into a clip of that person walking out on the wing of a plane in flight and landing a backflip, with 11 backdrops from a canyon to a night city.

FastH3 is a four-forward distilled MiniMax H3 model that generates synchronized video and stereo audio from a text prompt. This H200-backed Preview v1 checkpoint supports native 5.167–14.375 second T2AV clips at 24 FPS; FL2VA and Ref2VA were not distilled.

Run Unsloth’s GGUF quantizations of Qwen 3.8 Flash Next for long-context chat and reasoning. Includes optional thinking mode plus sampling and stop controls.

Refusal-reduced variant of Qwen 3.8 27B for chat and coding. It can emit or hide thinking traces and supports tight decoding controls. On Wiro a request takes about 25K input tokens and returns up to 16K.

Qwen3.8-27B is a 27B-parameter chat model from Qwen with optional thinking traces for coding, analysis, and agent workflows. On Wiro a request takes about 25K input tokens and returns up to 16K.

Chat model based on Qwen 3.8 27B with refusal behavior reduced through direction removal. On Wiro a request takes about 25K input tokens and returns up to 16K.

RMBG-2.0 removes image backgrounds with an 8-bit alpha matte for smooth edges. Built by BRIA AI for production cutouts trained on licensed data.

Create a short talking-portrait video from a human photo and a script. Choose an effect preset, aspect ratio, and 5, 10, or 15s duration.

A birthday video from a portrait in 11 scenes: nine styled as VIRGO magazine covers with a cake, a flash mirror selfie, and a tiara-and-sparklers party shot that needs a second photo.

Drops the person from one portrait into a flash-lit birthday scene and writes the age you enter into the shot, as number candles on a cake or glitter on a cheek.

Tries one of 24 hairstyles on a portrait, from fades and pompadours to bobs and waves, in a hair color you type. Face, clothes and background are kept from the photo.