SDXL Control-LoRA Canny
Structure-guided SDXL text-to-image using Stability AI Control-LoRA Canny (rank 256). Control-LoRA files load like ControlNets in ComfyUI/models/controlnet.
Estimated VRAM
~7.712 GB @ 1024x1024 — ⚠️ ESTIMATE (unverified real-world usage). Estimate, not a measured figure: weight-size floor = sd_xl_base_1.0.safetensors (6.938 GB) + control-lora-canny-rank256.safetensors (0.774 GB) = 7.712 GB, sizes verified on Hugging Face. Activations and the control image add more; ComfyUI offloading can reduce it. Source, checked 2026-10-14.
Models required
- sd_xl_base_1.0.safetensors — 6.938 GB ✅ verified 2026-10-14 (listing) · sha256
31e35c80fc4829d14f90153f4c74cd59c90b779f6afe05a74cd6120b893f7e5b→models/checkpoints - control-lora-canny-rank256.safetensors — 0.774 GB ✅ verified 2026-10-14 (listing) · sha256
21f79f7368eff07f57bcd507ca91c0fc89070d7da182960ff24ed1d58310c3a7→models/controlnet
Tested GPUs
No verified test reports yet.
Nodes
CheckpointLoaderSimple ControlNetLoader ControlNetApply CLIPTextEncode EmptyLatentImage LoadImage KSampler VAEDecode SaveImage
Common errors & fixes
- ControlNet has little/no effect / wrong conditioning image
- Pass a canny (or other matching) map, not a raw photo. Preprocess with a canny preprocessor node pack, or supply a pre-made edge map (per ComfyUI_examples ControlNet page). ✅ per source
- ControlNet model not found
- Place control-lora-canny-rank256.safetensors in ComfyUI/models/controlnet (Control-LoRAs use the same directory as ControlNets per source). ✅ per source
Low-VRAM tips
⚠️ General guidance, not individually verified.
- Use --lowvram
- Batch size 1 at 1024x1024
- Prefer Control-LoRA over full ControlNet (smaller, same folder)
Source & credit
Workflow by comfyanonymous (ComfyUI_examples) / Stability AI. Get it from the original: https://comfyanonymous.github.io/ComfyUI_examples/controlnet/. License: SDXL: CreativeML Open RAIL++-M; Control-LoRA: see Stability AI control-lora repo.