Last updated
Licences & credits
EditorsTab is built on other people's open-source work. Every model below was chosen on quality. Most are under permissive licences that allow commercial use of what you make; the ones marked ⚠️ are not, and say so.
AI models
These are the on-device packages (sizes are the one-time download; after that your browser keeps its own copy). The cloud runs the same model families at full precision — its list is in the engine's README.
| Model | Used for | Licence | Size |
|---|---|---|---|
| SiluetaSilueta (compressed U²-Net), ONNX export via the rembg project | Background removal (Standard) | MIT | 43 MB |
| BiRefNet_liteBiRefNet_lite (Peng Zheng et al.), fp16 ONNX export by onnx-community | Background removal (High) | MIT | 114.5 MB |
| RMBG-2.0RMBG-2.0 (BRIA AI), q4f16 ONNX export by BRIA AI | Background removal (Premium) | ⚠️ CC-BY-NC-4.0Non-commercial licence (BRIA) — a commercial launch needs a BRIA agreement or the High package. | 233.8 MB |
| MI-GANMI-GAN (Picsart AI Research) | Object removal / inpainting | MIT | 28.1 MB |
| Real-ESRGAN x4v3Real-ESRGAN (Xintao Wang et al.) | Upscaling (Standard) | BSD-3-Clause | 4.9 MB |
| Real-ESRGAN x4plusReal-ESRGAN (Xintao Wang et al.) | Upscaling (Premium) | BSD-3-Clause | 67.1 MB |
| DeOldifyDeOldify (Jason Antic) | Photo colorization (Standard) | MIT | 64 MB |
| DDColor-tinyDDColor (Xiaoyang Kang et al.), ONNX export by EditorsTab | Photo colorization (Premium) | Apache-2.0 | 135.3 MB |
| GFPGAN v1.4GFPGAN v1.4 (Xintao Wang et al.), ONNX via facefusion-assets, fp16 by EditorsTab | Face restoration (Standard) | Apache-2.0 | 170.3 MB |
| CodeFormerCodeFormer (Shangchen Zhou et al.), ONNX via facefusion-assets, fp16 by EditorsTab | Face restoration (Premium) | ⚠️ S-Lab-1.0S-Lab licence — non-commercial; a commercial launch uses the Standard package. | 188.7 MB |
| Informative Drawings (anime)Informative Drawings, anime style (Caroline Chan et al.), ONNX export by EditorsTab | Line-art sketch (Standard) | MIT | 17.2 MB |
| Informative Drawings (contour)Informative Drawings, contour style (Caroline Chan et al.), ONNX export by EditorsTab | Line-art sketch (Premium) | MIT | 17.2 MB |
| Informative Drawings (OpenSketch)Informative Drawings, OpenSketch style (Caroline Chan et al.), ONNX export by EditorsTab | Line-art sketch (Premium) | MIT | 17.2 MB |
| Selfie SegmenterMediaPipe Selfie Segmenter (Google) | Background blur | Apache-2.0 | 0.25 MB |
| BlazeFaceMediaPipe BlazeFace (Google) | Face detection (passport photos, restore) | Apache-2.0 | 0.23 MB |
| Selfie MulticlassMediaPipe Selfie Multiclass (Google) | Face cutout | Apache-2.0 | 16.4 MB |
| Kokoro-82MKokoro-82M (hexgrad), ONNX export by onnx-community | Text to speech | Apache-2.0 | 93 MB |
The restricted models
BRIA's RMBG-2.0 (Cutout Premium) and CodeFormer (Face Restore Premium) are the strongest models at their tasks and carry non-commercial licences. Each has a permissive package one tier down — Cutout High and Face Restore Standard — that gives results you may use commercially.
Face Swap runs in the cloud only, and two of the models behind it are also restricted: HyperSwap 1A (ResearchRAIL, research use only), which does the swap, and ArcFace w600k_r50 (InsightFace, non-commercial), which reads the face you supply. The rest of that pipeline — the BiSeNet face parser and the GFPGAN finish — is permissive.
The Vocal Separator runs Open-Unmix UMX-L, whose weights are CC BY-NC-SA 4.0 (non-commercial); its permissive sibling UMX-HQ (MIT) is one setting away. Every other model the cloud runs is under a permissive licence.
Attribution
Auto Captions runs Parakeet TDT 0.6B v3 by NVIDIA, licensed under CC BY 4.0, in an int8 ONNX conversion (istupakov/parakeet-tdt-0.6b-v3-onnx).
Libraries
- ONNX Runtime Web — MIT. Runs the image models on WebGPU or WebAssembly.
- MediaPipe Tasks Vision — Apache-2.0. Runs the portrait segmentation and face detection models.
- Transformers.js — Apache-2.0. Runs the speech model.
- kokoro-js — Apache-2.0. Text-to-speech pipeline around the Kokoro model.
- lamejs (@breezystack fork) — LGPL-3.0. Encodes MP3 audio in the browser. Used as an unmodified library and dynamically loaded, as the LGPL allows.
- Next.js — MIT. The web framework.
- React — MIT. The UI library.
- Tailwind CSS — MIT. The styling system.
- Lucide — ISC. The icon set.
Corrections
If you maintain something listed here and an attribution is wrong or missing, email hello@editorstab.com and it will be fixed.