Download OpenAPI specification:
Simple proxy service for vLLM that contains a plethora of premade tools for parsing and processing content of all kinds.
Currently hardcoded to listen at localhost:8080.
Outside of the listed responses, any upstream responses will currently be directly forwarded from vLLM -> your caller. Additionally, 400 responses are automatically generated from OpenAPI spec, similar to FastAPI.
If your instance has a key set, the header Authorization will be directly forwarded to "host" as-is.
Download the binary, run somewhere, and simply
http post -ef "http://localhost:8080/nsfw_check" --content-type application/json {
host: "http://localhost:7771/"
image: "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAEAAAABAAQAAAACCEkxzAAAADklEQVR42mOgAIyCUQAAAkAAAdJZOGAAAAAASUVORK5CYII="
} | table -e
Check image for NSFW content. This is a fine-grained approach that distinguishes multiple levels of explicit content. Current implementation is roughly based on the MPA film ratings
| Score | MPA | Description |
|---|---|---|
| ≈ 0.0 | NR | Irrelevant content |
| ≈ 0.2 | G | Safe content |
| ≈ 0.4 | PG | Suggestive Content |
| ≈ 0.6 | PG-13 | Partial/educational nudity |
| ≈ 0.8 | R | Nudity |
| ≈ 1.0 | NC-17 | Pornography |
| host required | string <uri> vLLM host to POST to |
| image required | string <uri> Image URL to be sent for classification. Can be a public URL or a base64 data URI scheme |
{
}
{- "usage": {
- "prompt_tokens": 374,
- "completion_tokens": 161,
- "prompt_tokens_details": {
- "cached_tokens": 1109
}
}, - "scale": 0.798
}Upsample prompts for generative inference. Specific behavior depends on the chosen style. Many models have their own unique styles optimized for the training distribution, so check there first if the performance is suboptimal!
| host required | string <uri> vLLM host to POST to | ||||||||||||||||||||
| prompt required | string Prompt to upsample | ||||||||||||||||||||
| style required | string Default: "default" Enum: "default" "ernie" "ernie_cn" "flux2" "flux2_img" "ideogram4" "krea2_img" "ltx2" "ltx2_img" Influences how the prompt will be upsampled. If your model is in the list, it's strongly recommended to use that!
| ||||||||||||||||||||
| image | string or null <uri> Image URL to be used as a reference for prompt generation. Can be a public URL or a base64 data URI scheme | ||||||||||||||||||||
| temperature required | number <double> [ 0 .. 3 ] Default: 1 Temperature used for sampling. Higher values result in more creative but may be less stable and hallucinate more. Use 1.0 to get the model's native distribution. | ||||||||||||||||||||
| reasoning required | boolean Default: false Enable reasoning when performing the prompt upsample. This will increase generation time and may reduce creativity, but could make the output more accurate to the inputs. |
{- "prompt": "photo of a cat holding a sign that reads 'BEAUTIFUL DAY'",
- "style": "ideogram4",
- "temperature": 1,
- "reasoning": false
}{- "usage": {
- "prompt_tokens": 374,
- "completion_tokens": 161,
- "prompt_tokens_details": {
- "cached_tokens": 1109
}
}, - "enhanced_prompt": "{\"high_level_description\":\"A photograph of a ginger tabby cat holding a small cardboard sign that reads 'BEAUTIFUL DAY'.\",\"compositional_deconstruction\":{\"background\":\"A sun-drenched wooden porch with weathered grey planks and a small patch of green clover grass visible in the upper-left corner, illuminated by bright midday sunlight.\",\"elements\":[{\"type\":\"obj\",\"desc\":\"A fluffy ginger tabby cat with white paws and a pink nose, sitting upright on its haunches, holding a small piece of brown corrugated cardboard between its two front paws.\"},{\"type\":\"text\",\"text\":\"BEAUTIFUL DAY\",\"desc\":\"Hand-written text in thick black permanent marker on the cardboard sign, centered with slightly uneven lettering, held at chest height by the cat.\"}]}}"
}