Cherry 2 Pro

Our flagship video model, with incredible cinematic motion. Clips up to 30 seconds at 1080p with audio, from text, up to 29 references or Characters, or a source video to edit.

Our newest and most capable image model. Lifelike light and skin, the sharpest small text, and edits that stay clean round after round. Keeps a character identical across a set and merges up to 10 references or Characters into one scene, at up to 2K.

The previous Mango, still a favorite for stylized art and rich fantasy worlds. Takes reference images and Characters, and renders up to a crisp 4K.

Our photorealism model at its very best, for portraits, fashion, and product shots with true-to-life skin, fabric, and light. Takes up to 3 references or Characters, up to 2K.

The Guava that started it all, kept for prompts tuned on it. Photoreal people and scenes up to 2K; Guava 2 is the current version.

OpenAI's quick, precise image model that follows the prompt to the letter and renders readable text. Generates or edits from up to 15 reference images, and makes animated GIFs. OpenAI's safety filter applies.

A fast text-to-image model with a beautiful built-in look, made for quick first passes. Outputs up to 2K; no reference images.

Google's image model for graphics and marketing content with legible text. Generates or edits from up to 14 reference images at up to 4K. Google's safety filter applies.

Blazing fast and easy on Gems. Photoreal text-to-image with solid anatomy and clean lettering, up to 2K; no reference images.

An open-weights text-to-image model that follows long, detailed prompts and renders English and Chinese text. Up to 2K with LoRAs; no reference images.

Our all-round video model. Start and end frames, up to 10 references or Characters, Voices and audio, and clips from 2 to 30 seconds at up to 1080p.

Pika Labs' six-step MiniMax-H3, made to follow timed shots and camera moves, with picture and stereo audio together. Start and end frames or up to 8 references (longer clips take fewer), 5 to 15 seconds at up to 768p.

Speech, dialogue, music, and sound effects from a single prompt, as an MP3 of exactly the length you ask, 5 seconds to 2 minutes. Match a voice from up to 3 audio clips.

ElevenLabs v4 text to speech in 16 voices and 90+ languages, directed line by line with audio tags like [whispers] or [laughs].

Eleven Music by ElevenLabs: songs with lyrics and vocals, or instrumentals, from a description, at exactly the length you pick, up to 5 minutes.

An anime and illustration model with clean linework and flat color. Text-to-image up to 2K, with Danbooru-style tags and prompt weighting.

An open-weights general image model on the Flux architecture. Text-to-image, with steps and guidance in your hands.

Flux 2 Dev, the open-weights model that thrives on long, detailed prompts. Generates or edits from up to 3 reference images, up to 2K.

xAI's Grok Imagine image model in two tiers. Generates or edits from up to 3 reference images while keeping the source's details intact. xAI's safety filter applies.

A quick general text-to-image model with a negative prompt. Great for everyday prompts, photoreal or illustrated.

Our high-detail SDXL pipeline with automatic face and hand cleanup. Text-to-image or from one input image; slower than plain SDXL, and worth it.

ByteDance's Seedream, from Seedream 5.0 Flash for drafts to Seedream 5.5, the newest and the default, with detailed scenes, readable text, and edits from up to 10 references or Characters, up to 4K on Seedream 4.5. Strict Filter: prompts and first images are checked before each run.

The original Stable Diffusion 1.5, at a budget price. Text-to-image or from one input image, perfect for matching older work.

The Stable Diffusion that follows detailed prompts most closely. Text-to-image or from one input image, at a budget price.

Classic Stable Diffusion XL, tuned for photoreal output. Text-to-image or from one input image, at a budget price.

Video that stays faithful to every reference you give it. A start frame, up to 9 references or Characters, or a source video to edit; 3 to 15 seconds at up to 1080p.

Set the first and last frame and watch it fill in the motion between, with audio. 2 to 15 seconds at 720p or 1080p.

xAI's Grok Imagine for video, from text, up to 5 references, or a short source clip. 1 to 15 seconds at up to 720p. xAI's safety filter applies.

Simple, dependable video from text or a start frame, with nothing else to configure. 5 or 10 seconds at up to 1080p.

A fast open-weights video model for cheap, rapid iteration. Text or a start frame, up to 720p, in Distilled and Dev variants.

Recasts a video you already have: send the clip and photos of who or what replaces it, and the motion, camera, timing and sound stay as filmed. Sources run 4 to 30 seconds.

Our anime video model, as Melon and Melon Pro. Start and end frames or up to 7 references or Characters, optional audio, 3 to 16 seconds at up to 1080p.

Full-feature video with audio on every clip and up to 2K output. Start and end frames, up to 9 references or Characters, and Voices; 4 to 15 seconds.

Character video from a start frame plus up to 5 references or Characters, with Voices and audio. 2 to 15 seconds at 720p or 1080p.

ByteDance's Seedance, from Seedance 2.0 Mini for drafts to Seedance 2.5, with clips up to 30 seconds, many references or Characters, and source-video edits. Strict Filter: the provider's content filter applies.

Alibaba's all-in-one Wan 3.0. Start and end frames, up to 10 references or Characters, Voices and audio, and clips from 2 to 30 seconds at up to 1080p. Strict Filter: the provider's content filter applies.

An open-weights video model, from text or a start frame, with a faster Lightning variant. Up to 720p.