Real-ESRGAN or SUPIR: which to use for upscaling photos
Different weight classes and, more importantly, different rights. Real-ESRGAN from Tencent ARC Lab is about 17M, runs even on a CPU, upscales photos 2–4x and cleans noise and compression artifacts; it is BSD-3-Clause, so commercial use has no strings. SUPIR is built on SDXL and additionally uses LLaVA 13B for captions, requires a GPU, and does not merely upscale but recreates detail in badly damaged images. However, SUPIR's license permits non-commercial use only, with commercial use requiring the authors' written permission, so it cannot go into a product without that agreement. Real-ESRGAN was last updated in 2022-04; SUPIR shipped in 2024-02.
Comparison based on catalog data
| Parameter | Real-ESRGAN | SUPIR |
|---|---|---|
| Category | Photo editing | Photo editing |
| Developer | Tencent ARC Lab, China | XPixel Group (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shanghai AI Lab and others), China |
| Releases | Jul 2021 – Apr 2022 | Feb 2024 – Feb 2024 |
| Sizes | about 17M | based on SDXL, plus LLaVA 13B for captions |
| Hardware | Laptop | 1 GPU |
| Commercial use | Commercial use allowed | Non-commercial only |
| License | BSD-3-Clause | Non-commercial use only (commercial use with the authors' written permission) |
| Russian | Not applicable | Not applicable |
| Ollama | No | No |
| Without GPU | Yes | No |
| Tasks |
|
|
Choose Real-ESRGAN if
- Your project is commercial: BSD-3-Clause imposes no conditions
- You need throughput: small product photos, artifact cleanup, print preparation
- You have no GPU and the model has to run on a CPU
Choose SUPIR if
- The image is damaged enough that detail must be recreated, not just scaled
- This is research, a pilot or an archive restoration with no commercial use
- You have a GPU and are ready to request written permission from the authors
Other comparisons
- BRIA RMBG or BiRefNet: which to use for background removal
- Qwen or GigaChat: which to choose for business
- Qwen or Llama: which to choose for business
- GigaChat or YandexGPT: which to choose for business
- DeepSeek or Qwen: which to choose for business
- Gemma or Llama: which to choose for business
- Mistral or Llama: which to choose for business
- Whisper or GigaAM: which to choose for business
- Whisper or Parakeet: which to choose for business
- Silero or Piper: which to choose for business
- FLUX or Stable Diffusion: which to choose for business
- FLUX or Qwen-Image: which to choose for business
Need a model for your task?
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
- SelectThe model and size for your task and hardware budget
- DeployOn your server or in a closed network, with an API
- Fine-tuneOn your data, or connect a knowledge base
- IntegrateInto your CRM, ERP, bot, website or team chat