BiRefNet Massive TR DIS5K TR TES
The 'trained on everything' cutout model
- image
- image
Of all the BiRefNet weight variants in this pack, Runware_birefnet_massive_tr_dis5k_tr_tes has the most intimidating name and the simplest story: this is the one that got trained on a lot of data. The "Massive TR" naming means it was trained with a large, merged training set - DIS5K and beyond - rather than a single benchmark. More training data, more domains, more subject types.
That makes it the closest thing to a "just throw any subject at it" option in the BiRefNet family. Real-world cutouts don't care which benchmark your image resembles; a massively-trained checkpoint has seen more of them.
How it works
Standard removeBackground contract - runware:112@8 over the Runware REST API, your image up as a data URI, cutout back as an IMAGE tensor. Same one-input surface as the other five BiRefNet nodes; only the checkpoint differs.
Inputs that matter
- image - the only input you set.
- outputFormat - PNG or WEBP or the alpha flattens in the default JPG. Repeated across this whole pack for a reason: it's the most common way to waste a run.
- outputQuality - 20–99, default 95.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt
Or ComfyUI Manager → search Runware → Install → restart. API key in ComfyUI Settings → Runware API key or RUNWARE_API_KEY.
Gotchas
- "Massive training" doesn't mean it beats the general model on every subject. The background-removal KB is blunt that nobody wins this outright - hair remains the universal failure case, and InSPyReNet still takes specific comparisons. It means the model generalizes across more domains, which is a good default when you have mixed subject matter.
- If you're cutting out one specific type of thing (fine edges, high-res detail), the targeted variants - Dis, HRSOD - will usually beat the jack-of-all-trades. Massive TR is the generalist's generalist.
The pick: when your batch has a messy mix of subjects and you want one checkpoint that handles all of them without thinking, Massive TR is the one. It's the "bring anything" cutout model.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| ttlopt | BOOLEAN | false | Enable to set ttl. Off uses the model's default. |
| ttl_valueopt | INT | 60 | Time-to-live (TTL) in seconds for generated content. Only applies when `outputType` is `URL`. |
| outputFormatopt | COMBO | JPG | File format for the generated image. |
| outputQualityopt | INT | 9520–99 | Compression quality of the output. Higher values preserve quality but increase file size. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |