Logo of Trellis2
Trellis 2
FeaturesPricing
BlogFAQ
Logo of Trellis2
Trellis 2

Transform images into stunning 3D models with AI-powered technology

Product

  • Features
  • Pricing
  • Blog
  • FAQ

Company

  • About
  • Contact
  • Trify3D

Legal

  • Privacy Policy
  • Terms of Service

© 2026 Trellis 2. All rights reserved.

Run TRELLIS 2 on 6GB/8GB VRAM: Low VRAM ComfyUI Guide (2026)
2026/04/21
3 min read

Run TRELLIS 2 on 6GB/8GB VRAM: Low VRAM ComfyUI Guide (2026)

How to run Microsoft TRELLIS 2 on low VRAM GPUs (RTX 3060, 4060) in ComfyUI. Covers low VRAM mode setup, resolution tuning, memory optimization, and performance benchmarks.

Last updated: April 21, 2026

You don't need a $2,000 GPU to run TRELLIS 2. With the right configuration, Microsoft's 4-billion parameter 3D generation model works on consumer GPUs with as little as 6GB VRAM. This guide shows you exactly how to optimize ComfyUI for low VRAM setups.

If your GPU doesn't meet the requirements, you can use TRELLIS 2 online instead — no GPU needed.

Which GPUs Work?

GPUVRAMMax ResolutionTime per Model
RTX 40608 GB1024³ (low VRAM)~120s
RTX 30608 GB1024³ (low VRAM)~150s
RTX 306012 GB1024³~90s
RTX 20606 GB512³ only~60s
GTX 10808 GB512³~45s
GTX 16606 GB512³~50s

Step 1: Enable Low VRAM Mode

In ComfyUI, launch with the low VRAM flag:

All Posts

Author

avatar for Trellis2 Team
Trellis2 Team

3D technology specialists focused on AI-powered 3D model generation, format conversion, and browser-based 3D rendering. We test and review 3D tools so you don't have to.

Categories

    Which GPUs Work?Step 1: Enable Low VRAM ModeStep 2: Configure TRELLIS Node ParametersStep 3: Optimize Your SystemClose GPU-Consuming ApplicationsUse fp16 PrecisionMonitor VRAM UsagePerformance BenchmarksRTX 3060 8GBRTX 4060 8GBTips for Best Results on Low VRAMTroubleshooting"CUDA out of memory" error"Generation is very slow""Output quality is poor"Related Guides

    More Posts

    Newsletter

    Join the community

    Subscribe to our newsletter for the latest news and updates

    python main.py --lowvram

    Or for extremely limited VRAM:

    python main.py --novram

    Step 2: Configure TRELLIS Node Parameters

    In the TRELLIS node settings within ComfyUI:

    ParameterValueWhy
    Resolution512Cuts VRAM usage by ~60%
    Low VRAM ModetrueEnables gradient checkpointing
    Batch Size1Process one model at a time
    Texture Resolution512Lower texture = less memory
    PBR Channelscolor onlySkip roughness/metallic/normal
    Steps20Fewer steps = less peak memory

    Step 3: Optimize Your System

    Close GPU-Consuming Applications

    Before running TRELLIS 2, close:

    • Browser tabs (especially with video/WebGL)
    • Other AI tools (Stable Diffusion, game engines)
    • GPU monitoring tools with live graphs
    • Screen recording software

    Use fp16 Precision

    # In ComfyUI startup
python main.py --lowvram --force-fp16

    This halves memory usage with minimal quality loss.

    Monitor VRAM Usage

    # Real-time GPU monitoring
watch -n 1 nvidia-smi

    Performance Benchmarks

    RTX 3060 8GB

    ResolutionLow VRAM ModeVRAM PeakTimeQuality
    512³Off7.5 GB~30sGood
    512³On6.2 GB~60sGood
    1024³On7.8 GB~180sVery Good

    RTX 4060 8GB

    ResolutionLow VRAM ModeVRAM PeakTimeQuality
    512³Off7.2 GB~20sGood
    1024³Off7.9 GB~100sVery Good
    1024³On6.8 GB~120sVery Good

    Tips for Best Results on Low VRAM

    1. Start at 512³ for iteration — Get the shape right, then switch to 1024³ for final output
    2. Use a fixed seed — Reproduce results without re-running
    3. Pre-process images — Resize to 512x512 before loading into ComfyUI
    4. Disable texture baking initially — Focus on shape quality first
    5. Process one at a time — Don't queue multiple models

    Troubleshooting

    "CUDA out of memory" error

    • Reduce resolution to 512³
    • Enable --lowvram or --novram
    • Close all other applications
    • Set texture resolution to 256

    "Generation is very slow"

    Low VRAM mode trades speed for memory. On a 6GB card, expect 2-3x longer generation times compared to a 24GB card.

    "Output quality is poor"

    • Ensure input image is high quality (512x512+)
    • Use 1024³ for final output (not 512³)
    • Increase steps to 30-40

    Related Guides

    • TRELLIS 2 ComfyUI Guide — Full installation instructions
    • ComfyUI Workflow Tutorial — Step-by-step workflow
    • TRELLIS 2 Download — Model weights and dependencies
    How to Use TRELLIS 2 Online Free: No GPU Required (2026)

    How to Use TRELLIS 2 Online Free: No GPU Required (2026)

    Complete guide to using Microsoft TRELLIS 2 online for free. Covers Hugging Face Spaces, Google Colab, and other cloud platforms — no GPU or Python installation needed.

    avatar for Trellis2 Team
    Trellis2 Team
    2026/05/04
    Image to 3D Model for 3D Printing: Complete Workflow Guide (2026)

    Image to 3D Model for 3D Printing: Complete Workflow Guide (2026)

    Turn any photo into a 3D-printable model. Covers AI tool selection, mesh repair in Blender, STL export settings, slicer configuration, and troubleshooting common print failures. Works with TRELLIS 2, Meshy AI, and Tripo AI.

    avatar for Trellis2 Team
    Trellis2 Team
    2026/05/03
    TRELLIS 2 PBR Texture Generation in ComfyUI: Complete Guide (2026)

    TRELLIS 2 PBR Texture Generation in ComfyUI: Complete Guide (2026)

    How to generate production-quality PBR textures (Base Color, Roughness, Metallic, Normal maps) with TRELLIS 2 in ComfyUI. Covers texture baking nodes, parameter tuning, and export for Unity/Unreal.

    avatar for Trellis2 Team
    Trellis2 Team
    2026/04/21