Generate Low Poly Characters on Only 6GB VRAM (Free & Local)
Creating low poly characters for your games is now possible locally on your own PC using Trellis 2 and Pixel 3D — no cloud subscription required. This guide walks you through the entire workflow: generating a character, making it low poly, texturing, rigging, animating, and importing into your preferred software. Best of all, you only need 6GB of VRAM to run the whole process.
What You’ll Learn
By the end of this guide, you’ll be able to:
- Install and set up Trellis 2 & Pixel 3D in ComfyUI
- Generate low poly characters from 2D images
- Fix common issues like background artifacts and orientation
- Optimize quality with full models (if you have higher VRAM)
- Rig and animate your character using free tools
Prerequisites
Before starting, ensure your system meets these requirements:
- NVIDIA RTX GPU with at least 6–8GB VRAM (tested on Windows only)
- ComfyUI installed (see our ComfyUI Installation Guide for help)
- Enough disk space for models and outputs
If your hardware isn't powerful enough or you want to speed up generations, consider using RunPod's cloud GPU service.
Special Offer - $5 Credit Included!
When you sign up for RunPod using our affiliate link, you'll receive a $5 credit that can be used to generate up to 9,000 images and 300 videos. This gives you plenty of resources to explore ComfyUI and AI image/video generation without any upfront cost!
Linux & Mac Users
The installation instructions in this guide are for Windows only. For Linux or Mac, check the official Trellis 2 documentation linked in the video description.
Step-by-Step Process
Step 1: Install Trellis 2 & Pixel 3D via ComfyUI Easy Install
The easiest way to get everything running is to use the ComfyUI Easy Install package developed by Iivo. This package bundles all dependencies, so you don’t need to manually handle Python packages.
- Download the ComfyUI Easy Install package from the link provided in the video description (pixel-artistry.com/trellis2installationguide).
- Extract the folder to a location with plenty of disk space.
- Launch ComfyUI: Inside the extracted folder, run start_comfyui_flash_attention.bat.
- ComfyUI will open in your browser automatically.
Why Easy Install?
The Easy Install package is specifically designed for 3D workflows. It pre-configures everything for Trellis 2 and Pixel 3D, saving you from dependency headaches.
Pro Tip
If you encounter issues, check the common installation FAQ section on the download page. The Iivo tool also includes a debug mode.
Step 2: Load the Low Poly Workflow
Once ComfyUI is running, you need to open the provided workflow.
- Click the "Open" button in the top-left corner of ComfyUI.
- Select the workflow file named pixel_artistry_pixelart_for_trellises_to_low_poly (download it from the video description).
- Examine the workflow layout: You’ll see several sections:
- Image input (top) – where you upload your character image
- Generation section (left) – controls for low poly settings
- Texturing section (right) – textures the low poly mesh based on the reference image
Important Compatibility Note
The mesh texturing node only works with Trellis 2, not Pixel 3D. Ensure you select Trellis 2 when texturing. Pixel 3D can be used for the initial mesh generation only.
Step 3: Upload Your Image and Configure Settings
The workflow expects a square (1:1 ratio) image with a solid background. This works best with Trellis 2.
- Upload your character image by dragging it into the Load Image node at the top.
- Set the low poly face count – the Low Poly Face Number parameter. Use 5,000 for a good balance of detail and polygon count.
- Choose the model in the Trellis 2 Low Poly GGUF node. Options include:
- Trellis 2 GGUF – recommended for 6GB VRAM
- Pixel 3D GGUF – great for props (e.g., rocks, buildings)
- For models inside: Q4, Q5, Q6, Q8 – Q8 gives the highest quality but uses more VRAM.
- Adjust generation steps (e.g., 25 for mid quality, 50 for best).
- Optionally increase resolution for higher quality.
- Texture steps – keep at 25 for a good result; higher values increase quality but cost more memory.
Full Models vs GGUF
The workflow includes alternative nodes for users with 12GB+ VRAM. You can enable them with Ctrl+B. These use the full (uncompressed) models for better quality. If you have enough VRAM, this is recommended for sharper textures and meshes.
Pro Tip
Always experiment with different seeds and face counts. Start with 5K faces and 25 steps to get a feel for generation speed. Adjust upward if you have time and VRAM.
Step 4: Generate the Low Poly Character
Now you’re ready to generate.
- Click the "Run" button (or press Ctrl+Enter) to start the queue.
- Wait for completion – generation typically takes 2–3 minutes per character.
- Review the results: The workflow outputs:
- Base mesh (untextured) – you can view it in the viewer node
- Textured mesh – the final low poly character with textures
Common Issue: Blue Spots
You may notice blue artifacts (especially on the back). This is caused by the built-in background remover not being accurate. See the next step for a fix.
First Character Generated!
You now have a low poly character ready for refinement. Even if the texture has minor flaws, the front side is usually very good.
Step 5: Fix Background Issues (Blue Spots)
To eliminate blue spots, use an external background removal tool and replace the built-in remover.
- Use a separate background remover – the transcript mentions Zimage Turbo 2K upscaler (a free ComfyUI custom node) to upscale and remove the background.
- Upload your original image to that tool and remove the background. Save the result as a PNG with alpha channel.
- In ComfyUI, replace the original image with your new alpha-PNG.
- Disable the built-in "Remove Background" node (set it to bypass or remove it).
- Reconnect the image output of your background remover node to the Image with Alpha input of the main workflow.
- Re-run the workflow.
What If Orientation is Wrong?
Sometimes the orientation of the mesh is flipped (e.g., upside down). To fix this:
- Find the Mesh with Box to Try Mesh node
- Set the Reorient Vertices option to None
- Re-run the generation This corrects the orientation without affecting geometry.
Pro Tip
For better results, you can regenerate the texture with a different seed after fixing the background. This often improves texturing especially on complex areas like weapons.
Step 6: Generate Props and Batch Assets
The same workflow works excellently for props (rocks, buildings, etc.) using the Pixel 3D GGUF model.
- Upload a prop image (e.g., a stone or building) with a solid background.
- Switch the model to Pixel 3D GGUF in the generation node.
- Run the workflow – you’ll get a low poly prop with good texture quality.
- For batch generation: You can queue multiple images by connecting a batch node. Generation time per asset remains ~2–3 minutes.
Examples from the Video
The creator demonstrated a stone (Lux to climb) and a house building – both produced clean low poly results with decent textures. Background props for mobile games are a great use case.
Step 7: Use Full Models for Higher Quality (12GB+ VRAM)
If you have 12GB or more VRAM, you can enable the high‑VRAM nodes for significantly better results.
- Enable the "big nodes" at the top of the workflow using Ctrl+B.
- Increase the low poly face number to 10,000 for a smoother mesh.
- Select the full (non‑GGUF) models – these are uncompressed and use the model’s full power.
- Run the workflow – you’ll get a more detailed shape and improved texture quality, especially on the front side.
Pro Tip
Even with 6GB VRAM, you can still use the GGUF models for decent results. Only upgrade to full models if you have the VRAM headroom.
Step 8: Rig and Animate Your Character
Once the character is generated, you can import it into Blender and use free rigging tools.
- Locate your output files: All 3D models are saved in the ComfyUI output folder.
- Import into Blender (or any 3D software that supports GLB).
- Export as GLB – make sure to merge vertices before export to clean up the mesh.
- Use Mesh to Motion (free local tool) for rigging and animation:
- Import the GLB into Mesh to Motion
- Choose skeleton size and align body parts (arms, shoulders, fingers)
- Select an animation from the library
- Alternative: Mixamo – the free online tool:
- Upload your GLB to Mixamo
- Select a running or walking animation
- Enable "In Place" to keep the character stationary
- Download the animated character and import back into Blender
Animation Quality
The animation quality depends on the original mesh. Characters with complex geometry (e.g., a sword on the back) may require manual cleanup. For best results, generate the character without separate props (like wings or swords) and add them later in Blender.
Final Animated Character!
You now have a completely rigged and animated low poly character, ready for use in a game engine or further refinement.
Related Guides
- ComfyUI Installation Guide – Complete setup of ComfyUI, Manager, and custom nodes
- Running ComfyUI on RunPod – Run this workflow in the cloud if your PC lacks VRAM
- Deploy Your First RunPod – Basic account and pod deployment
Next Steps
Now that you’ve generated and animated your first low poly character:
- Experiment with different image inputs – try characters with clean silhouettes for best results
- Optimize the workflow – reduce face count for mobile games or increase for PC
- Explore Pixel 3D GGUF for props and environment assets – they work exceptionally well
- Check out Mesh to Motion for advanced rigging features
- Use Mixamo for quick animations without local tools
For more advanced workflows, consider integrating 3D AI Studio’s Flow (node-based canvas for 3D) to split, edit, and retopologize models directly in the browser.