MIRAGA · Public service information
Understand how each feature works, what you can upload, and which models and access providers are used, without signing in.
Select a character and type text messages. When creating a character, provide a name, background, personality and greeting, and upload an avatar image. You can replace the avatar when editing. A real-person photo or video is not required.
Characters appear as static avatars with written profiles. Avatars can be cartoon, illustrated or realistic; a real-person photo is allowed only when owned by that person or explicitly authorized. The model uses character settings and conversation context to generate text replies for fictional interaction and storytelling.
Uploading an avatar does not animate a real person or mean that person is chatting with you. This feature does not produce real-person video or clone their voice. Characters are AI simulations, not real people, endorsements or professional advisers. Impersonation, sexual/NSFW interaction and non-consensual intimate content are prohibited.
Text-chat model: Doubao Seed 1.6. Access provider: fal.ai. Chat access costs 59 credits for 7 days, without automatic renewal.
Provide a text prompt and optionally upload reference images you own or are authorized to use. Supports text-to-image, reference-based generation and editing.
Outputs AI-generated or edited static images, including illustrated, cartoon or realistic styles, subject to the AUP.
Provide a text prompt and, where supported, reference images or first/last frames. Select duration, resolution and available audio options.
Outputs AI-generated video; models with audio support can produce a synthetic soundtrack. Real-person material must not be used for impersonation or NCII.
Provide a script or instructions, or upload PDF, TXT, Markdown or reference audio. Reference audio can be transcribed free and used for translation, rewriting and voice-over.
Seed Audio 1.0 generates voice-over, translated speech, multi-speaker podcasts and speeches, with scene generation and iterative revisions. Reference transcription uses Doubao Speech ASR; both are accessed through fal.ai. Uploading a voice is not authorization to clone or impersonate its speaker.
Mira is a Miraga interface name, not a claim that we independently trained a model of that name. Mira 2.0 uses Seedream; other options use Qwen Image / Qwen Image Edit, Wan and the models mapped below. All listed models are accessed through fal.ai.
| Feature / interface name | Underlying model | Access provider |
|---|---|---|
| Mira / Mira 2.5 / Mira 2.5 fast | Qwen Image / Qwen Image Edit | fal.ai |
| Mira 2.0 | Seedream | fal.ai |
| Mira 1.5 | Qwen Image Edit 2511 | fal.ai |
| Mira 2.1 | Wan 2.7 Image | fal.ai |
| GPT Image 2 (gpt2) | GPT Image 2 | fal.ai |
| Banana 2 | Nano Banana | fal.ai |
| Ref Lite / Ref Lite Back | Seedance 2.0 Mini | fal.ai |
| Ref Plus / Ref Plus Back | Seedance 2.0 Fast | fal.ai |
| Ref Pro | Seedance 2.5 | fal.ai |
| HD Ref Plus | Berry 1.0 | fal.ai |
| HD Audio | Wan 2.7 | fal.ai |
| Mira Ref HD | Wan 3.0 | fal.ai |
| Voice-over / translation / podcast / speech | Seed Audio 1.0 | fal.ai |
| Reference audio transcription | Doubao Speech ASR | fal.ai |
| AI Companion Chat / AI Character Chat | Doubao Seed 1.6 | fal.ai |
Model availability and versions may change. This disclosure does not claim endorsement of Miraga by any model developer. Use is also subject to applicable model/provider licenses and content restrictions.
I confirm that I own this material or have the rights holder’s explicit permission to upload and use it for AI processing. For real people’s photos, video or voice, I have their informed consent. I will not impersonate anyone or create non-consensual intimate content.