What is Miaohua
Miaohua is launched by SenseTimeAI painting creation tool, based on the 7-billion-parameter Artist large model, users can quickly generate high-quality paintings through text prompts or uploading images. It supports multiple styles and precise control functions, and can custom-train LoRA models to adapt to specific styles. The platform supports automatic completion of Chinese prompts, can import open-source models from communities such as Hugging Face and Civitai, provides drag-and-drop LoRA training, and can customize personalized IPs, digital portraits, and artistic styles. The Miaohua platform simplifies the creation process and is suitable for multiple fields such as design, marketing, and game animation, helping users quickly turn ideas into visual content.
Main features of Miaohua
-
AI online drawing: Enter text or upload a reference image to quickly generate ultra-high-definition paintings, with support for automatic completion of Chinese prompts.
-
Open-source model import: One-click import of community models from Hugging Face, Civitai, GitHub, etc., eliminating local deployment costs.
-
LoRA training: Upload images and use drag-and-drop operations to train exclusive models such as personalized IPs, digital portraits, and artistic styles.
-
Inspiration Square: Browse and like others' works, and recreate the same style with one click to spark creative inspiration.
-
ControlNet control: Precisely control image generation through pose control, line art coloring, depth detection, and more.
-
HD and image expansion: Supports 6K HD image generation, as well as 1.25x-2.0x multi-ratio image expansion and partial repainting.
How to use Miaohua
- Visit the official website: Visit the Miaohua official website https://miaohua.sensetime.com/, and click the login button.
- Enter drawing: Click "AI online drawing" at the top to enter the creation page.
- Enter a description: Enter the image description in the prompt box. Chinese is supported, and the automatic completion function can be used.
- Upload a reference image(optional): If you need image-to-image, upload a reference image to assist generation.
- Select model and style: Choose the self-developed Artist model or an imported open-source model, and select your preferred artistic style.
- Set parameters: Adjust parameters such as image size and generation quantity.
- Start generating: Click the generate button and wait a few seconds to get the painting.
- Download or recreate: If satisfied, download and save it; you can also continue optimizing with partial repainting and image expansion.
- Advanced options: Recreate the same style with one click in "Inspiration Square", or upload images to train an exclusive LoRA model.
Core advantages of Miaohua
-
"Self-developed + open-source" dual-track model ecosystem: It has both the self-developed Artist large model and one-click import of community models from Hugging Face, Civitai, and GitHub.
-
Extremely low LoRA training threshold: Drag-and-drop can train personalized IP, digital portrait, and artistic style models, with no need for a local graphics card or deployment knowledge.
-
Chinese-native + Prompt completion: Good understanding of Chinese prompts; a small number of prompts is enough for automatic completion, making it beginner-friendly.
-
Computing power acceleration: Powered by SenseTime's self-developed SenseCore computing cluster, images that take 10 seconds locally can be generated in the cloud in about 2 seconds.
Comparison of SenseMirage with similar competing products
| Dimension | SenseMirage(SenseTime) | Dreamina AI(ByteDance) |
|---|---|---|
| Underlying model | Self-developed Artist v1.0 Alpha (tens-of-billions-parameter Transformer diffusion model) | Seedream 5.0 series(Preview released in February, Pro version released on July 8) |
| Killer feature | Open-source model ecosystem + LoRA training | Web-search-based image generation + deep thinking + precise brush editing |
| Free quota | 10 images per day + invitation bonuses,no watermark | 60-66 credits per day (about 3 credits per image), free versionforced watermark |
| Maximum resolution | 6K high definition | 2K direct output + 4K enhancement |
| Image editing | Local repainting, image expansion (1.25x-2.0x) | Box selection + natural language local modification, multi-turn conversational editing |
| Text rendering | Average | Precise rendering of dense text in 10+ languages, Chinese text accuracy 92%+, can directly produce posters |
| Multi-image consistency | Relies on LoRA training | General reference system: fuses up to 14 reference images, maintaining character consistency without fine-tuning |
| Open-source model import | ✅ One-click import from Hugging Face / Civitai / GitHub | ❌ Closed ecosystem |
| LoRA training | ✅ Drag-and-drop cloud training, low barrier to entry | Relies on built-in capabilities, weak professional customization |
Application scenarios of SenseMirage
- Creative design: helps designers quickly realize the transformation from concept to visual image, suitable for fashion, advertising and various other design needs.
- Digital marketing: provides marketers with eye-catching visual materials to enhance promotional effects.
- Games and animation: generates game characters and animation scenes with a sense of dynamism and storytelling.
- Architectural visualization: assists architects and engineers in presenting architectural design proposals, improving communication efficiency.
