Skip to content

docs: add 1-click swm cloud workspace links to README - #181

Open
itssutch wants to merge 1 commit into
microsoft:mainfrom
itssutch:docs/swm-cloud-workspace
Open

itssutch wants to merge 1 commit into
microsoft:mainfrom
itssutch:docs/swm-cloud-workspace

Conversation

@itssutch

Copy link
Copy Markdown

Closes #180

Adds a 1-click swm cloud badge and installation links to README.md for launching MoGe-2 and MoGe-3 in ComfyUI on a cloud GPU with pre-loaded weights and starter workflows:

  1. Vision, Matting & PBR Decomposition Studio (BiRefNet + Lucida + Marigold v2 + MoGe) (https://cloud.swmgpu.com)
  2. SAM 3.1 Video Multiplex Tracking & SpEdE 3D Gaussian Splat Object Editor (https://cloud.swmgpu.com)
  3. ComfyArt Infinite Canvas, 360° Panorama Stickers & Qwen-Image-Layered Studio (https://cloud.swmgpu.com)

Workspace 1: vision-pbr-matte-studio

  • Minimum VRAM: 24 GB VRAM (verified on NVIDIA RTX 3090 24 GB; ~4.3 GB peak VRAM for MoGe / BiRefNet / Lucida; 82 GB min / 100 GB suggested volume; runs from $0.07/hr)
  • Pre-loaded Hugging Face Model Weights (~29.11 GB total):
    • Comfy-Org/MoGe (1484985, MIT): moge_2_vitl_normal_fp16.safetensors (661.9 MB), moge_3_vitl_fp16.safetensors (740.7 MB)
    • Comfy-Org/BiRefNet (884af70, MIT): birefnet.safetensors (444.5 MB), lucida.safetensors (884.9 MB)
    • Comfy-Org/marigold-v2-0 (c304e43, Apache-2.0): qwen_image_edit_2509_int8_convrot.safetensors (20.50 GB) + albedo/depth/normals adapters & VAEs (3 × 1.70 GB + 3 × 253.8 MB + 3 × ~2.5 MB)
  • Starter Workflows: 03_MoGe_monocular_depth_geometry.json, 01_BiRefNet_and_Lucida_background_removal.json, 02_Marigold_v2_albedo_depth_normals_PBR.json

Workspace 2: sam3-1-spede-3d-splat-editor

  • Minimum VRAM: 24 GB VRAM (verified on RTX Pro 4000 24 GB; 90 GB min / 125 GB suggested volume; runs from $0.07/hr)
  • Pre-loaded Hugging Face Model Weights (~27.69 GB total):
    • Comfy-Org/MoGe (1484985, MIT): moge_2_vitl_normal_fp16.safetensors (661.9 MB), moge_3_vitl_fp16.safetensors (740.7 MB)
    • Comfy-Org/sam3.1 (1.75 GB), Comfy-Org/Qwen-Image-2.1 (7.26 GB + 7.26 GB + 9.35 GB + 675.5 MB)
  • Starter Workflows: 02-spede-3d-splat-object-remove.json, 03-spede-3d-splat-object-replace.json, 01-sam3-1-multiplex-video-segmentation.json

Workspace 3: comfyart-panorama-canvas-studio

  • Minimum VRAM: 48 GB VRAM (verified on NVIDIA A40 48 GB; 145 GB min / 175 GB suggested volume; runs from $0.30/hr)
  • Pre-loaded Hugging Face Model Weights (~52.41 GB total):
    • Comfy-Org/MoGe (1484985, MIT): moge_2_vitl_normal_fp16.safetensors (661.9 MB), moge_3_vitl_fp16.safetensors (740.7 MB)
    • Comfy-Org/Qwen-Image-2.1 (7.26 GB + 9.35 GB + 675.5 MB), Comfy-Org/Qwen-Image-Layered (20.53 GB + 253.8 MB) & Comfy-Org/Qwen-Image (9.38 GB), Comfy-Org/sam3.1 (1.75 GB), Comfy-Org/BiRefNet (444.5 MB + 884.9 MB), gribbliest/qwen-image-2.1-qma-camera-angles (318.8 MB), nomadoor/qwen-image-2.1-360-panorama-outpaint-lora (159.4 MB)
  • Starter Workflows: 04-panorama-moge3-anyangle.json, 01-panorama-stickers-360-scene.json, 02-comfyart-camera-angles-canvas.json, 03-qwen-image-layered-decomposition.json

All three workspaces support Run (ephemeral clean snapshot sessions) and Fork (persistent customizable workspace volumes), direct SSH access, and MCP agent control (https://api.swmgpu.com/api/mcp) built on open-source swm-gpu.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1-click cloud GPU ComfyUI workspaces for MoGe on swm cloud

1 participant