🎨 Mtrini-Imagine-1.0-7B

🧠 An image-generation experiment by Compiwer AI.

✨ Building AI For Everyone.

Welcome to Mtrini-Imagine-1.0-7B — Compiwer AI's first Mtrini image-generation release! 🚀

This repository contains the LoRA Adapter version of Mtrini-Imagine.

Instead of shipping the entire model, this release contains the learned Mtrini-Imagine updates that can be applied to the compatible HiDream-O1-Image base model.

That means a tiny ~98 MB adapter can bring the Mtrini-Imagine modifications to the base model. 🪄

---

🌟 What's inside?

🔧	Details
🤖 Model	Mtrini-Imagine-1.0-7B
🧩 Type	LoRA Adapter
🏗️ Base	HiDream-O1-Image-Dev
🧠 Architecture	HiDream-O1 / UiT
⚡ Framework	DiffSynth-Studio
📐 LoRA Rank	16
🔢 LoRA Alpha	32
🧮 Precision	BF16
🖼️ Training Examples	4,993
💾 Adapter Size	~98 MB

---

🎨 What is Mtrini-Imagine?

Mtrini-Imagine is Compiwer AI's experiment in open image generation.

The goal is simple:

💡 Take powerful open image-generation architectures and experiment with making them our own.

Mtrini-Imagine 1.0 was trained as a LoRA adapter on top of HiDream-O1-Image.

The result is a compact adapter that can be distributed separately from the massive base model. 📦

---

🧪 Training

Mtrini-Imagine-1.0-7B was trained using DiffSynth-Studio with a LoRA-based training setup.

⚙️ Configuration

• 🏗️ Base: HiDream-O1-Image
• 🧩 Method: LoRA
• 📐 Rank: 16
• 🔢 Alpha: 32
• 🧮 Precision: BF16
• 🖼️ Examples: 4,993
• 🛠️ Framework: DiffSynth-Studio

The adapter contains the learned parameter updates rather than a complete copy of the original model.

---

🚀 Why use the Adapter?

The adapter is perfect if you:

• 🧑‍💻 Already have HiDream-O1-Image
• 💾 Want a much smaller download
• 🧪 Want to experiment with LoRA
• 🔧 Work directly with DiffSynth-Studio
• 🧠 Want to study the Mtrini-Imagine modifications

📦 Download size

~98 MB

Compared with the full model, that's tiny. 😎

---

🛠️ Usage

Load the compatible HiDream-O1-Image base model and apply the Mtrini-Imagine adapter using DiffSynth-Studio.

import torch

from diffsynth import ModelConfig
from diffsynth.pipelines.hidream_o1_image import HiDreamO1ImagePipeline

pipe = HiDreamO1ImagePipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(
            path="PATH_TO_HIDREAM_MODEL"
        )
    ]
)

pipe.load_lora(
    pipe.dit,
    "Mtrini-Imagine-1.0-7B.safetensors",
    alpha=1.0
)

⚠️ Exact loading configuration can vary between DiffSynth-Studio versions.

---

🧠 Model Capabilities

Mtrini-Imagine is designed for experimentation with:

🎨 Creative image generation
🖼️ Text-to-image generation
🧪 Image-model research
🤖 Multimodal AI experimentation
💻 Local inference
🔬 Open model development

---

⚠️ Known Limitations

Like basically every image model, Mtrini-Imagine isn't perfect. 😅

You may encounter:

• ✍️ Incorrect text inside images
• 🔤 Misspelled words
• 🧩 Strange object relationships
• 🎨 Unexpected visual artifacts
• 🧠 Prompt interpretation errors
• 🔍 Inconsistent small details

Text rendering can be especially challenging.

We're continuing to experiment with future Mtrini-Imagine versions. 🚀

---

🛡️ Safety

Please use Mtrini-Imagine responsibly.

Do not use the model to create illegal, harmful, deceptive, or abusive content.

Users are responsible for ensuring their generated content follows applicable laws and platform policies.

---

📜 License

This adapter is released under Apache-2.0, subject to the licensing conditions of the underlying HiDream-O1-Image model.

Please review the original model's license before commercial use.

---

❤️ Credits

Built by Compiwer AI 🇲🇦

Model family: Mtrini
Model: Mtrini-Imagine-1.0-7B
Base: HiDream-O1-Image
Framework: DiffSynth-Studio

---

🔗 Explore

🏠 Compiwer AI
🤗 More Mtrini models
🧪 More experiments coming soon

---

🚀 Mtrini-Imagine-1.0-7B

🎨 Imagine more. Build more. Ship more.

Compiwer AI — Building AI For Everyone.