3DMDT1: Generalist SD 2.1 Checkpoint for 3D Renders

TL;DR

• SD 2.1 768 checkpoint optimized for 3D render aesthetics (Blender/Unreal look)
• Trigger word: 3dmdt1 — add photography keywords for realism, "3d render" for 3D style
• Recommended: DPM++ 2M Karras, 30 steps, CFG 7
• ~2,500 downloads, 485 likes on Civitai

What Is 3DMDT1?

3DMDT1 is a fine-tuned Stable Diffusion 2.1 768 checkpoint created by ArtificialGuyBR (with collaboration from Shadow_Shinigami of the PRMJ model). The project began as a dedicated 3D-oriented model — the goal was to generate images that look like they were rendered in Blender, Unreal Engine, Octane, or similar 3D tools. Through iterative merging and refinement, it evolved into a generalist checkpoint that also handles photorealism, fantasy illustration, landscapes, and portraits.

The model is distributed as fp16 SafeTensor (2.4 GB) and fp16 .ckpt (2.4 GB), each with a YAML config file. It was published on Civitai in February 2023 and has accumulated ~2,500 downloads and 485 likes.

Trigger Words & Prompting

The model uses a single trigger word: 3dmdt1. Include it in your prompt to activate the model's trained style.

For 3D render output (the model's default tendency):

• Include keywords: 3d render, octane render, blender, unreal engine, volumetric lighting, subsurface scattering

For photorealistic output:

• Add photography terms: raw photo, photograph, 8k uhd, dslr, sharp focus, film grain

For illustration/fantasy:

• Use: illustration, fantasy art, digital painting, concept art

The author notes: "Because the model is more focused on 3d rendering, if you don't insert anything it will create something more 3d related than photorealistic."

Recommended Generation Settings

From the model card (all gallery images used these):

• Sampler: DPM++ 2M Karras
• Steps: 30
• CFG Scale: 7
• Negative prompt: Standard (author uses default)
• Embeddings: None required
• Face restore: Off
• Resolution: 768×768 (native), can upscale

File Formats & Installation

Format | Size | Notes

|--------|------|-------

3DMDT1-FP32-NoPruned-Safe.safetensors | 2.4 GB | fp16 SafeTensor, recommended 3DMDT1.ckpt | 2.4 GB | fp16 legacy checkpoint 3DMDT1.yaml | 1.77 KB | Config file — place alongside model Download from Civitai, place the .safetensors (or .ckpt) and .yaml in your models/Stable-diffusion folder. In Automatic1111/WebUI, refresh the model list and select 3DMDT1.

Common Errors & Fixes

Error | Cause | Fix | Source

|-------|-------|-----|--------

Model loads but outputs noise | Missing config YAML | Download 3DMDT1.yaml and place next to model file | Civitai model page Black/white images | Wrong precision for .ckpt | Use SafeTensor version or ensure fp16 support | Civitai model page Overly 3D look when wanting realism | Missing photography keywords | Add raw photo, photograph, 8k to prompt | Model card

FAQ

Q: What base model is 3DMDT1 trained on? A: Stable Diffusion 2.1 768 (SD 2.1 at 768×768 resolution).

Q: Does it work with SDXL or SD 1.5? A: No, it's an SD 2.1 768 checkpoint. Use with SD 2.1-compatible pipelines only.

Q: What's the trigger word? A: 3dmdt1 — include it in every prompt.

Q: Is there a SafeTensor version? A: Yes, 3DMDT1-FP32-NoPruned-Safe.safetensors (2.4 GB) is the recommended format.

Q: Who created this model? A: ArtificialGuyBR, with collaboration from Shadow_Shinigami (PRMJ), Mousewrites, PePPa, Queria Star Morta, and PublicPrompts.

Sources

• Civitai model page — model files, description, stats, sample gallery
• ArtificialGuyBR portfolio — author's other models and projects

Related Models

• PRMJ by Shadow_Shinigami — collaborative base for 3DMDT1
• Other ArtificialGuyBR checkpoints — explore the full model collection
• SD 2.1 768 base model — the foundation this fine-tune builds on