Can My GPU Run IP-Adapter FaceID SDXL?
Image to Image · VRAM requirements and compatibility for 78 GPUs
Home › AI Models › IP-Adapter FaceID SDXL
Preserve face identity across generated images. IP-Adapter FaceID SDXL is a image to image model — developed by Tencent, released 2024-01, 22M parameters, 98 MB download, license: apache-2.0.
Short answer: you need at least 6 GB of VRAM (10 GB recommended). 78 of 78 GPUs can run it; the smallest NVIDIA card that can is the GTX 1060 6GB (6 GB).
Check IP-Adapter FaceID SDXL on your GPUIP-Adapter FaceID SDXL VRAM requirements
| Format | VRAM |
|---|---|
| FP16 (full precision) | 10 GB |
| Minimum | 6 GB |
| Recommended | 10 GB |
| Optimal | 16 GB |
Tip: Face identity preservation. Requires InsightFace alongside.
Available as: ComfyUI, Diffusers, CPU offloading.
Official pages: Hugging Face
Which GPUs can run IP-Adapter FaceID SDXL?
Ratings go from S (perfect) to F (can't run) and use the best format that fits in each card's VRAM. How ratings work.
NVIDIA
| GPU | VRAM | Rating | Runs as | Speed | |
|---|---|---|---|---|---|
| RTX 5090 | 32 GB | Perfect | FP16 | Fast | Details |
| RTX 5080 | 16 GB | Great | FP16 | Fast | Details |
| RTX 5070 Ti | 16 GB | Great | FP16 | Fast | Details |
| RTX 5070 | 12 GB | Good | FP16 | Fast | Details |
| RTX 5060 Ti 16GB | 16 GB | Good | FP16 | Fast | Details |
| RTX 5060 Ti 8GB | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 5060 | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 4090 | 24 GB | Perfect | FP16 | Fast | Details |
| RTX 4080 Super | 16 GB | Great | FP16 | Fast | Details |
| RTX 4080 | 16 GB | Great | FP16 | Fast | Details |
| RTX 4070 Ti Super | 16 GB | Great | FP16 | Fast | Details |
| RTX 4070 Ti | 12 GB | Good | FP16 | Fast | Details |
| RTX 4070 Super | 12 GB | Good | FP16 | Fast | Details |
| RTX 4070 | 12 GB | Good | FP16 | Moderate | Details |
| RTX 4060 Ti 16GB | 16 GB | Good | FP16 | Moderate | Details |
| RTX 4060 Ti 8GB | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 4060 | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 3090 Ti | 24 GB | Great | FP16 | Fast | Details |
| RTX 3090 | 24 GB | Great | FP16 | Fast | Details |
| RTX 3080 Ti | 12 GB | Good | FP16 | Fast | Details |
| RTX 3080 12GB | 12 GB | Good | FP16 | Fast | Details |
| RTX 3080 10GB | 10 GB | Usable | FP16 | Moderate | Details |
| RTX 3070 Ti | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 3070 | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 3060 Ti | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 5050 | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 3060 | 12 GB | Good | FP16 | Moderate | Details |
| RTX 2080 Ti | 11 GB | Usable | FP16 | Moderate | Details |
| RTX 2080 Super | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 2080 | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 2070 Super | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 2070 | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 2060 Super | 8 GB | Usable | Offload | Very Slow | Details |
| RTX 2060 | 6 GB | Barely | Offload | Very Slow | Details |
| GTX 1080 Ti | 11 GB | Usable | FP16 | Moderate | Details |
| GTX 1080 | 8 GB | Usable | Offload | Very Slow | Details |
| GTX 1070 Ti | 8 GB | Usable | Offload | Very Slow | Details |
| GTX 1070 | 8 GB | Usable | Offload | Very Slow | Details |
| GTX 1060 6GB | 6 GB | Barely | Offload | Very Slow | Details |
AMD
| GPU | VRAM | Rating | Runs as | Speed | |
|---|---|---|---|---|---|
| RX 9070 XT | 16 GB | Great | FP16 | Fast | Details |
| RX 9070 | 16 GB | Good | FP16 | Fast | Details |
| RX 7900 XTX | 24 GB | Great | FP16 | Fast | Details |
| RX 9070 GRE | 12 GB | Good | FP16 | Moderate | Details |
| RX 7900 XT | 20 GB | Great | FP16 | Fast | Details |
| RX 9060 XT 16GB | 16 GB | Good | FP16 | Moderate | Details |
| RX 9060 XT 8GB | 8 GB | Usable | Offload | Very Slow | Details |
| RX 7900 GRE | 16 GB | Good | FP16 | Moderate | Details |
| RX 7800 XT | 16 GB | Good | FP16 | Moderate | Details |
| RX 7700 XT | 12 GB | Good | FP16 | Moderate | Details |
| RX 7600 XT | 16 GB | Good | FP16 | Moderate | Details |
| RX 7600 | 8 GB | Usable | Offload | Very Slow | Details |
| RX 6950 XT | 16 GB | Good | FP16 | Moderate | Details |
| RX 6900 XT | 16 GB | Good | FP16 | Moderate | Details |
| RX 6800 XT | 16 GB | Good | FP16 | Moderate | Details |
| RX 6800 | 16 GB | Good | FP16 | Moderate | Details |
| RX 6700 XT | 12 GB | Usable | FP16 | Moderate | Details |
| RX 6600 XT | 8 GB | Barely | Offload | Very Slow | Details |
| RX 6600 | 8 GB | Barely | Offload | Very Slow | Details |
Apple
| GPU | VRAM | Rating | Runs as | Speed | |
|---|---|---|---|---|---|
| M5 Ultra | 96 GB | Usable | FP16 | Fast | Details |
| M3 Ultra | 96 GB | Usable | FP16 | Fast | Details |
| M4 Max | 48 GB | Usable | FP16 | Fast | Details |
| M5 Max | 48 GB | Usable | FP16 | Fast | Details |
| M4 Pro | 24 GB | Barely | FP16 | Moderate | Details |
| M5 Pro | 24 GB | Barely | FP16 | Moderate | Details |
| M4 | 16 GB | Barely | FP16 | Slow | Details |
| M3 Max | 36 GB | Usable | FP16 | Fast | Details |
| M3 Pro | 18 GB | Barely | FP16 | Moderate | Details |
| M5 | 16 GB | Barely | FP16 | Slow | Details |
| M6 | 16 GB | Barely | FP16 | Moderate | Details |
| M3 | 16 GB | Barely | FP16 | Slow | Details |
| M2 Ultra | 192 GB | Usable | FP16 | Fast | Details |
| M2 Max | 96 GB | Usable | FP16 | Fast | Details |
| M2 Pro | 32 GB | Barely | FP16 | Moderate | Details |
| M2 | 24 GB | Barely | FP16 | Moderate | Details |
| M1 Ultra | 128 GB | Usable | FP16 | Fast | Details |
| M1 Max | 64 GB | Usable | FP16 | Fast | Details |
| M1 Pro | 32 GB | Barely | FP16 | Moderate | Details |
| M1 | 16 GB | Barely | FP16 | Slow | Details |
Frequently asked questions
How much VRAM does IP-Adapter FaceID SDXL need?
IP-Adapter FaceID SDXL needs at least 6 GB of VRAM, 10 GB recommended, and 10 GB to run at full FP16 precision.
Can IP-Adapter FaceID SDXL run on 8 GB of VRAM (e.g. NVIDIA RTX 4060)?
On the NVIDIA RTX 4060: yes, with CPU offloading (rating C, usable).
Can IP-Adapter FaceID SDXL run on the NVIDIA RTX 3060 (12 GB)?
Yes, in FP16 (rating B, good).
Which GPUs can run IP-Adapter FaceID SDXL?
78 of the 78 GPUs we track can run it, and 32 run it well (rating B or better). The cheapest one that runs it well is the NVIDIA RTX 3060 (12 GB, about $329).
Does IP-Adapter FaceID SDXL work on AMD or Apple Silicon?
19 AMD Radeon and 20 Apple Silicon chips in our list can run it. There is no MPS backend, so it may not work on Apple Silicon.