Gaussian Splat to Image · Image to Image

ABOUT THE WORKFLOW

Restore a Gaussian Splat Render
Upload a screenshot from any 3D Gaussian Splat viewer. The workflow reads the perspective, color, and depth cues from the render and fills in blank areas, removes floating artifacts, and restores the scene into a clean photorealistic image. A specialized LoRA trained on Gaussian Splat restoration handles the conversion. The output is a clean image at approximately 1 megapixel.

Model

• Qwen Image Edit 2511 by Alibaba. An image editing model paired with a Gaussian Splat restoration LoRA that converts raw 3DGS renders into clean scene images, and a Lightning LoRA for 4-step generation.

HOW IT WORKS

Step 1. Upload your Gaussian Splat screenshot
A screenshot or render from any Gaussian Splat viewer: SuperSplat, the SHARP preview, Unity, Unreal, Blender, or a web viewer. The image can have blank areas, floating splats, and noise.
Works great with: SuperSplat exports · SHARP previews · Unity 3DGS renders · web viewer screenshots

Step 2. Edit the prompt (optional)
The default prompt instructs the model to restore the scene perspective and repair blank areas. Leave it as-is for most cases. Edit it to steer the restoration toward a specific look or to describe what the missing areas should contain.

Step 3. Hit run and download
Qwen Image Edit 2511 processes the Gaussian render in 4 steps and returns a clean photorealistic image.
Ready for: Photoshop · Figma · presentations · social media · portfolio work

First time? Upload a Gaussian Splat screenshot and hit run. The prompt and settings are preconfigured.

RECOMMENDED SETTINGS

Quick-start guide. Find the goal that matches yours and copy the settings.

• Standard Gaussian restoration — Default prompt, 4 steps, CFG 1, seed increment. Upload your screenshot and run.

• Large blank areas in the render — Add context to the prompt describing what the missing areas should contain. "Restore the scene. The missing areas are a continuation of the forest floor and sky" gives the model direction for filling large gaps.

• More detail in the restored image — Increase the total pixel count from 1 megapixel toward 1.6 megapixels. Sharper output, longer generation.

• Result looks overprocessed or painterly — Lower the Gaussian restoration LoRA strength from 1.0 toward 0.6. The model applies less aggressive cleanup, keeping more of the original render's texture.

• Floating artifacts persist in the output — The source screenshot may have too many stray splats for a single pass. Crop the worst areas out of the screenshot before uploading, or run the output through the workflow a second time.

• Want a different style instead of photorealism — Rewrite the prompt to describe the target look. "Restore the scene as a watercolor painting" or "Restore the scene as a clean architectural rendering" shifts the aesthetic.

Prompt: The default prompt handles most Gaussian Splat screenshots. Edit it only when blank areas need specific content described. "Restore the perspective of the scene and fill the missing areas with a continuation of the brick wall and cobblestone street" is specific. "Fix it" is not.

LEARN

📹 Videos

• Intro to Floyo

• ComfyUI 101 Free Course ft. Sebastian Kamph

• Floyo 101 for Team Collaboration

✨ Quick links

• Help docs

• LoRA Training guide

• Discover more workflows

USE CASES

🏗️ Clean Up SHARP or TripoSplat Output
Generate a Gaussian Splat from a single image with SHARP, screenshot a novel angle from the 3D preview, and restore it into a clean photograph with this workflow.

🎮 Convert Game Engine 3DGS Renders to Clean Stills
Take a screenshot from a Unity or Unreal Gaussian Splat scene and turn it into a polished image for marketing materials, documentation, or portfolio work.

📐 Fill Missing Areas in Photogrammetry Captures
Photogrammetry and 3DGS captures often have holes and sparse regions. Screenshot the problem angle and restore it into a complete scene image.

🖼️ Generate Novel Views from 3D Splats
Orbit a Gaussian Splat to a new camera angle in the viewer, screenshot it, and restore the render into a clean image at that perspective.

WHAT WORKS BEST / WHAT TO AVOID

✅ Works great

• Gaussian Splat screenshots with visible scene content and clear perspective cues

• Renders with moderate blank areas (under 40% of the frame)

• Screenshots from well-reconstructed splats with dense coverage in the center

• Single-scene images with consistent lighting

⚠️ May produce softer results

• Screenshots that are mostly blank with minimal scene content

• Renders with extreme floating artifacts covering the subject

• Very low-resolution viewer screenshots (under 512px)

• Multiple disconnected scene fragments in one screenshot

FAQ

What is Gaussian Splat to Image restoration?
Gaussian Splat to Image takes a raw screenshot from a 3D Gaussian Splat viewer and converts it into a clean, photorealistic image. The model reads the perspective, color, and depth information in the render and fills in blank areas, removes floating splat artifacts, and restores surface detail. The result looks like a photograph taken from that camera angle rather than a 3DGS render.

What Gaussian Splat viewers can I screenshot from?
Any viewer that displays 3D Gaussian Splats: SuperSplat, the ComfyUI GeomPack preview, Unity with a 3DGS plugin, Unreal Engine, Blender, or any web-based PLY viewer. Take a screenshot at the angle you want restored and upload it.

Does this workflow work with the SHARP Image to 3D workflow?
Yes. Generate a Gaussian Splat with SHARP from one photo, orbit the 3D preview to a new angle, screenshot it, and feed that screenshot into this workflow. The output is a clean image of the scene from the new camera angle.

Can I control what fills the blank areas in the restoration?
Yes. Edit the prompt to describe what the missing regions should contain. The model uses both the visible splat content and your text description to fill gaps. More specific prompts produce more accurate fills.

What is the Gaussian restoration LoRA?
A specialized LoRA adapter trained on pairs of Gaussian Splat renders and their corresponding clean images. It teaches the model to recognize splat artifacts, floating points, and blank regions, and replace them with clean scene content that matches the perspective and lighting of the original render.

Is Qwen Image Edit 2511 licensed for commercial use?
Qwen Image Edit 2511 is an open-source model by Alibaba. The Gaussian restoration LoRA and Lightning LoRA have their own license terms. Check each component's license on its model page for commercial use in your specific project.

How to run Gaussian Splat to Image restoration online?
You can run Gaussian Splat to Image restoration online through Floyo. No installation, no setup, no local GPU needed. Open the workflow in your browser, upload your inputs, and hit run. Free to try.

WHY FLOYO?

Floyo is the only platform with team collaboration for ComfyUI in the browser. You run workflows with no install. You share run history, assets, and models across your team. You pay only when you generate. Floyo supports open-source and closed-source models.

A designer runs an edit and likes the result. A teammate opens that exact run from shared history and keeps going. No file handoffs. No version confusion.

For studios and enterprise teams, Floyo adds private workspaces, pooled resources, and a team usage dashboard. Other ComfyUI cloud tools run for one person at a time. Floyo runs for the whole team, with transparent per-generation costs.

Ready to try it?
Upload a Gaussian Splat screenshot and hit run. The restoration prompt and LoRA are already configured.

→ Launch Workflow, Free

Questions? Watch the free course or check the FAQ above.