Youchuan API for Native 2K Stills and Motion

The Youchuan API gives developers programmatic access to Youchuan, Midjourney's official Chinese platform, through a single Atlas Cloud key. It covers six generation modes across two model generations, V8.1 and V8.2, spanning text-to-image, image-to-image, image-to-video, multi-image blending, style transfer, and background removal. Billing is pay-as-you-go per call, so there is no subscription to sign before your first request.

Six modes ship in the family, and each one is available on both V8.1 and V8.2. Text-to-image, image-to-image, and blend return four candidates per run, style transfer repaints an input while holding its composition, image-to-video animates a still into four five-second clips, and background removal returns one transparent cutout. Blend accepts two to five source images in a single call.

V8.1 shipped on April 30, 2026 as the fastest model in the lineup, roughly four to five times quicker than earlier versions, and it introduced native 2K rendering through the hd flag. V8.2 followed on July 27, 2026 with an aesthetics and image quality pass: renders skew bolder and more refined, and low-quality outliers appear far less often. Both versions remain callable, so you can pin V8.1 for consistency with existing work or move to V8.2 for the newer look.

Create an Atlas Cloud account, generate one API key, then send a request naming the model ID for the mode and version you want, such as youchuan/v8.2/text-to-image. Rates are transparent and charged per call, and Day-0 access means each new version is callable as soon as it lands. Start building today.

Image modes default to 1:1 and support nine aspect ratios, from square through cinematic 21:9 and tall 9:21. Setting hd to true produces native 2K output instead of an upscaling pass, billed at 1.5x the base rate. On the motion side, image-to-video returns four five-second clips at either 480p or 720p, with motion set to low or high.

Prompt-driven modes expose the controls Midjourney users already know: aspect_ratio, stylize from 0 to 1000, chaos up to 100, weird up to 3000, and quality at either 1 or 4. Prompts accept up to 1024 characters, and an optional sref image URL steers the aesthetic without dictating composition. Style transfer and background removal are deliberately simpler, taking an input image plus a target style prompt for the former.

Generation is asynchronous. Your call returns a request_id, and polling the prediction endpoint reports status as created, processing, completed, or failed, with the outputs array holding result URLs once the job finishes. Set enable_base64_output to true when you would rather receive the payload inline than fetch hosted files.

Most modes on the Youchuan API run $0.086 per call, covering text-to-image, image-to-image, blend, background removal, and image-to-video, while style transfer is priced at $0.129 per call. Two multipliers sit on top of those rates: hd images bill at 1.5x the base rate, and 720p video bills at 3x the 480p rate. Nothing is bundled into a plan, so you pay only for the calls you make.

Yes. Fix seed to a specific integer and the same prompt with the same settings reproduces that render exactly, whereas the default value of -1 randomizes every run. For a shared look across a campaign, pass one sref reference image with a steady stylize value and vary only the prompt.

Output shapes are fixed: clips run five seconds, blend takes at most five images, and prompts cut off at 1024 characters. Style transfer exposes no aspect ratio, hd, or stylize parameters, so resolution and framing follow the input image. Since jobs are queued and polled rather than streamed, build status checks and retries into your pipeline instead of expecting a synchronous response.