You can deploy this model and start earning money today!

MapleAndJoker/
emo

Views

No views yet

SD3 S2A Single-Repo API

This folder is a single-repo deployment template for your SD3 Stage2A model.

It is designed for this setup:

upload your API code
upload your Stage2A adapter files
do not upload the official SD3 base model

The base model stays external and is referenced by repo id:

stabilityai/stable-diffusion-3.5-medium

Repo layout

text

sd3_s2a_hf_single_repo_code/
  model/
    pytorch_lora_weights.safetensors
    emotion_encoder.pt
    inference_config.json
  s2a_sd3_api/
    modeling.py
    runner.py
    server.py
  tools/
    export_into_model_dir.py
  requirements.txt
  pyproject.toml
  README.md

What you upload

Upload this whole folder to one Hugging Face repo, but keep the official base model out of it.

Your repo should contain:

API code in s2a_sd3_api/
adapter files in model/
install metadata: requirements.txt, pyproject.toml

It should not contain:

the full SD3 base weights
training checkpoints unrelated to inference

Fill the `model/` directory

From your training checkpoint, copy these files into model/:

pytorch_lora_weights.safetensors
emotion_encoder.pt
inference_config.json

You can do that with:

bash

python tools/export_into_model_dir.py \
  --src_model_dir /path/to/sd3-stage-a-emotion/checkpoint-500

This writes directly into:

./model/

Install

bash

pip install -r requirements.txt
pip install -e .

Run locally

By default, the server loads adapter files from:

./model

bash

python -m s2a_sd3_api.server \
  --pretrained_model_name_or_path stabilityai/stable-diffusion-3.5-medium \
  --host 0.0.0.0 \
  --port 8000

If your model files live somewhere else, override it:

bash

python -m s2a_sd3_api.server \
  --pretrained_model_name_or_path stabilityai/stable-diffusion-3.5-medium \
  --s2a_model_dir /path/to/adapter_dir \
  --host 0.0.0.0 \
  --port 8000

API

Health

curl http://127.0.0.1:8000/health

Generate image

bash

curl -X POST http://127.0.0.1:8000/generate \
  -H "Content-Type: application/json" \
  -d '{
    "scene_prompt": "A quiet beach with a small wooden boat near the shore",
    "va": [-0.7, 0.7],
    "seed": 42
  }' \
  --output out.png

Generate JSON with base64 image

bash

curl -X POST http://127.0.0.1:8000/generate_json \
  -H "Content-Type: application/json" \
  -d '{
    "scene_prompt": "A quiet beach with a small wooden boat near the shore",
    "valence": -0.7,
    "arousal": 0.7,
    "seed": 42
  }'

Notes

This server handles one generation at a time per process.
For concurrent serving, run multiple workers behind a reverse proxy.
Users still need access to the official base model repo.

MapleAndJoker/emo

SD3 S2A Single-Repo API

Repo layout

What you upload

Fill the model/ directory

Install

Run locally

API

Health

Generate image

Generate JSON with base64 image

Notes

MapleAndJoker/
emo

Fill the `model/` directory