Yumina

Guides

What is Yumina Atelier? The AI image, video and 3D studio inside Yumina (2026)

Atelier makes pictures, video clips, 3D models, voices and music for the characters in your AI roleplay worlds. Here is each studio, the models behind it, and what a run costs.

Short answer, as of 15 November 2026: Atelier is the generation studio inside Yumina, an open-source platform for AI roleplay and AI games. It turns a sentence or a picture into images, video clips, 3D models, moving characters, voices and music. Each result is saved to your asset library, so it can go into a world you are building. It has six studios: Image, Video, Audio, 3D model, Live character and LoRA training.

What can you make in Yumina Atelier?

Studio What goes in What comes out Price of one run
Image a sentence, with or without a reference picture pictures from 512 px to 4K, depending on the model about 35 mushies on the default model
Video a sentence, a first frame, a last frame or character references a clip of 1 to 30 seconds at 480p to 4K, depending on the model 60 mushies for 5 seconds at 480p on Wan 2.2
3D model one picture of one object a textured .glb file 360 to 1,440 mushies by model, 630 on the default
3D model, Moving person one full-body front picture a rigged .glb with 10 motions 960 mushies
Live character one upper-body picture of a character a layered 2D character that blinks, breathes and follows the pointer 1,500 mushies
LoRA training one portrait, or 8 to 40 pictures a LoRA that keeps one character or one art style the same 2,300 mushies from one portrait
Audio a line of text or lyrics voice lines, songs, instrumental music, sound effects shown on the button

Atelier also has tools that work on a finished picture or clip: remove background, upscale 2x or 4x, inpaint, expand, relight, lip sync, add sound to a video, dub a video into 16 languages, and motion transfer.

Which AI models does Atelier offer?

For images, the "Describe & create" mode lists 30 models from other companies. They include Seedream, Nano Banana, GPT Image, FLUX, Qwen Image, Grok Imagine, Ideogram, Recraft and Krea. The default is Seedream 5.0 Lite. The "Custom generation" mode runs on Yumina's own GPUs instead. It has five art styles (Anime, Realistic, 2.5D, Ink Wash, Western), takes up to 3 LoRAs per picture, and exposes steps, CFG, sampler and seed.

For video, the model list holds 35 models in 13 families: Seedance, Kling, Veo, Gemini Omni, Wan, Hailuo, Grok Imagine, HappyHorse, LTX, Runway, Marey, FLUX Video and P-Video. The picker shows the ones available when you open it. How to make AI video of your character has the length and resolution of each.

For 3D, you choose between Hunyuan 3D 3.1 Pro, Rodin Gen-2.5, Tripo H3.1, TRELLIS.2 and Meshy 7.1. Songs come from ElevenLabs, ACE-Step or Mureka.

What does Yumina Atelier cost?

Every run is paid in mushies, Yumina's credits, and the Generate button shows the price before you press it. Yumina's docs say every account gets free mushies every month. Mushie packs start at $2.99 for 1,000.

The docs give these rules for pictures. The default model costs about 35 mushies per image, charged by actual usage once the image arrives. If no image comes back, you are not charged. Up to 2 pictures generate at the same time, and you can submit up to 30 requests an hour. A LoRA training that fails or is cancelled is refunded in full.

Yumina is open source, and its help page says "we don't make money from how much you chat". The vision page gives the reason: profiting from AI usage pushes a platform toward "offering worse models, shortening context windows, and selling user data". For the story itself you can bring your own key from a provider such as OpenRouter and pay that provider directly.

How does Atelier connect to worlds and characters on Yumina?

Everything Atelier makes lands in your asset library, and a card (Yumina's word for a world or character you publish) can use it from there.

  • A picture becomes a character's portrait, a scene image, a cover or a picture inside an opening.
  • A video has a "Put it in…" menu with three places: a character's moving portrait, the top of the opening, or the player screen.
  • A LoRA attaches to a character under "Illustration LoRA", so story illustrations draw that character with the same face every time. It can also set the art style of the whole card, or go on your own persona.
  • A 3D model goes into a 3D scene. Tell the Creation assistant "put a brass oil lantern on the bar counter" and it draws the object, turns the picture into a model and places it. 3D characters in a scene idle and turn to look at the player.
  • A voice reads a character's lines.

Finished work can also be shared. A recipe saves the model, settings and prompt of something you made, so another person can reuse it. The Plaza on the Discover page has a shelf each for pictures, videos, music, sound effects and voices.

Where is Atelier in the app?

There are three ways in: the Atelier card on the Create page, the Atelier button in the top right of My Library → Assets, and the Atelier section inside the editor. Signed out, you can open the studios and read the settings. Generating needs an account. Atelier runs on yumina.io. The open-source edition you host yourself does not include it.

If you have not built a world yet, How to make your own AI text adventure covers the canvas the pictures go into.

FAQ

Is Yumina Atelier free?

Each run costs mushies, and the button shows the price first. Every Yumina account gets free mushies every month. The docs list no separate free allowance for image or video generation.

Is Yumina Atelier like Higgsfield?

Both put many image and video models in one place. Atelier adds 3D models, LoRA training and a direct path from the result into a playable world. Higgsfield alternatives compares them on price and free use.

Can I use my own LoRA in Atelier?

Yes. Paste a Civitai link, upload a .safetensors file, pick one from the LoRA square, or train one from a single portrait. How to keep an AI character consistent with a LoRA has the steps.

What file does the 3D studio produce?

A .glb file with textures. You can turn it around on the page, download it, or place it in a 3D scene. How to turn a character image into a 3D model lists the five models and their prices.

Sources

Checked 15 November 2026. Image generation rules, the 35-mushie default, the limits and illustration LoRAs: Visuals and audio. LoRAs for players: Playing a world. Free mushies and the open-source statement: Is Yumina free?. Model families: Which AI models can I use?. The quoted reason: the vision page. 3D, video, LoRA and recipe features: the update log at yumina.io/app/updates, entries of 3 October, 10 November, 14 November and 15 November 2026. Model lists, run prices and pack prices: the Atelier model pickers and price settings as shipped that day.

Worlds you can play now

See every world on Yumina

More on atelier

Read next