Advanced ComfyUI workflow for Stable Projectorz — One-click 4K game asset pipeline
I spent a long time trying to get Stable Projectorz working with ComfyUI — couldn't find working solutions anywhere. Finally got it working AND pushed it to the next level. Sharing the full workflow for free. Hope it helps the community!
- ✅ Stable Projectorz bridge (Depth ControlNet)
- ✅ Tile ControlNet for upscale detail preservation
- ✅ SDXL + Refiner (two-stage sampling)
- ✅ LoRA Stack (4 simultaneous LoRAs)
- ✅ IP-Adapter for style/color transfer
- ✅ Ultimate SD Upscale (2-pass, up to 4K)
- ✅ IC-Light for post-process lighting control
- ✅ DSINE Normal Map generation
- ✅ VAE Decode Tiled
- ✅ ImageSharpen + ColorCorrect
| Node | Link |
|---|---|
| ComfyUI-StableProjectorzBridge | github.com/tianlang0704/ComfyUI-StableProjectorzBridge |
| ComfyUI_UltimateSDUpscale | github.com/ssitu/ComfyUI_UltimateSDUpscale |
| ComfyUI-Impact-Pack | github.com/ltdrdata/ComfyUI-Impact-Pack |
| ComfyUI_IPAdapter_plus | github.com/cubiq/ComfyUI_IPAdapter_plus |
| comfyui-ic-light | github.com/kijai/ComfyUI-IC-Light |
| comfyui_controlnet_aux | github.com/Fannovel16/comfyui_controlnet_aux |
| ComfyUI-post-processing-nodes | github.com/EllangoK/ComfyUI-post-processing-nodes |
| rgthree-comfy | github.com/rgthree/rgthree-comfy |
- Any SDXL base model (tested with WildcardXL Lightning)
sd_xl_refiner_1.0.safetensors4x-ClearRealityV1.pth→models/upscale_models/controlnet-tile-sdxl(TTPlanet V2 recommended)control-lora-depth-rank256.safetensorsip-adapter-plus_sdxl_vit-h.safetensors→models/ipadapter/CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors→models/clip_vision/iclight_sd15_fc.safetensors→models/IC-Light/dreamshaper_8.safetensors(for IC-Light)
Stable Projectorz (3D Model)
↓ Depth ControlNet
SDXL KSampler (base, steps 0→10)
↓
SDXL Refiner KSampler (steps 10→14)
↓
VAE Decode (Tiled)
↓
Tile ControlNet + Ultimate SD Upscale (pass 1, x2)
↓
Ultimate SD Upscale (pass 2, x2)
↓
ImageSharpen → ColorCorrect
↓
Projectorz Output (back to Stable Projectorz)
↓ (optional)
IC-Light → DSINE Normal Map
-
Bypass Tile ControlNet for faster generation during testing
-
Bypass IC-Light and Normal Map if not needed
-
IP-Adapter weight 0.8+ for strong style transfer
-
For characters: bypass Depth ControlNet, use OpenPose instead
-
SDXL Lightning models: keep CFG at 2.0-3.0, steps 12-14
All default settings in this workflow are optimized for WildcardXL Lightning model. If you use a different SDXL model, you may need to adjust:
- CFG (Lightning models use 2.0-3.0, standard SDXL uses 5.0-8.0)
- Steps (Lightning: 12-14, standard SDXL: 20-30)
- Sampler/Scheduler settings
- Eular/sgm_uniform or simple
For the Depth ControlNet, control-lora-depth-rank256.safetensors is recommended.
Other SDXL depth models may work but results may vary.
Special thanks to Igor (Stable Projectorz developer https://stableprojectorz.com/ ) and the ComfyUI community.
Workflow developed with assistance from Claude Sonnet 4.6 (Anthropic).
